Professional Documents
Culture Documents
Published titles
The Cambridge Handbook of Phonology, edited by Paul de Lacy
The Cambridge Handbook of Linguistic Code-switching, edited by Barbara E. Bullock
and Almeida Jacqueline Toribio
The Cambridge Handbook of Child Language, Second Edition, edited by Edith L. Bavin
and Letitia Naigles
The Cambridge Handbook of Endangered Languages, edited by Peter K. Austin and
Julia Sallabank
The Cambridge Handbook of Sociolinguistics, edited by Rajend Mesthrie
The Cambridge Handbook of Pragmatics, edited by Keith Allan and Kasia M. Jaszczolt
The Cambridge Handbook of Language Policy, edited by Bernard Spolsky
The Cambridge Handbook of Second Language Acquisition, edited by Julia
Herschensohn and Martha Young-Scholten
The Cambridge Handbook of Biolinguistics, edited by Cedric Boeckx and Kleanthes K.
Grohmann
The Cambridge Handbook of Generative Syntax, edited by Marcel den Dikken
The Cambridge Handbook of Communication Disorders, edited by Louise Cummings
The Cambridge Handbook of Stylistics, edited by Peter Stockwell and Sara Whiteley
The Cambridge Handbook of Linguistic Anthropology, edited by N. J. Enfield, Paul
Kockelman and Jack Sidnell
The Cambridge Handbook of English Corpus Linguistics, edited by Douglas Biber and
Randi Reppen
The Cambridge Handbook of Bilingual Processing, edited by John W. Schwieter
The Cambridge Handbook of Learner Corpus Research, edited by Sylviane Granger,
Gaëtanelle Gilquin and Fanny Meunier
The Cambridge Handbook of Linguistic Multicompetence, edited by Li Wei and Vivian
Cook
The Cambridge Handbook of English Historical Linguistics, edited by Merja Kytö and
Päivi Pahta
The Cambridge Handbook of Formal Semantics, edited by Maria Aloni and Paul
Dekker
The Cambridge Handbook of Morphology, edited by Andrew Hippisley and Greg
Stump
The Cambridge Handbook of Historical Syntax, edited by Adam Ledgeway and Ian
Roberts
The Cambridge Handbook of Linguistic Typology, edited by Alexandra Y. Aikhenvald
and R. M. W. Dixon
The Cambridge Handbook of Areal Linguistics, edited by Raymond Hickey
The Cambridge Handbook of Cognitive Linguistics, edited by Barbara Dancygier
The Cambridge Handbook of Japanese Linguistics, edited by Yoko Hasegawa
The Cambridge Handbook of Spanish Linguistics, edited by Kimberly L. Geeslin
The Cambridge Handbook of Bilingualism, edited by Annick De Houwer and Lourdes
Ortega
www.cambridge.org
Information on this title: www.cambridge.org/9781107130456
DOI: 10.1017/9781316418086
© Cambridge University Press & Assessment 2023
This publication is in copyright. Subject to statutory exception and to the provisions
of relevant collective licensing agreements, no reproduction of any part may take
place without the written permission of Cambridge University Press & Assessment.
First published 2023
A catalogue record for this publication is available from the British Library.
Library of Congress Cataloging-in-Publication Data
Names: Bentley, Delia, editor. | Mairal Usón, Ricardo, 1965– editor. | Nakamura,
Wataru, editor. | Van Valin, Robert D., Jr., editor.
Title: The Cambridge handbook of role and reference grammar / edited by Delia
Bentley, Ricardo Mairal Usón, Wataru Nakamura, Robert D. Van Valin, Jr.
Description: Cambridge, United Kingdom ; New York, NY : Cambridge University
Press, 2023. | Series: Cambridge handbooks in language and linguistics |
Includes bibliographical references.
Identifiers: LCCN 2022056326 (print) | LCCN 2022056327 (ebook) |
ISBN 9781107130456 (hardback) | ISBN 9781107571440 (paperback) |
ISBN 9781316418086 (epub)
Subjects: LCSH: Role and reference grammar. | LCGFT: Essays.
Classification: LCC P166 .C36 2023 (print) | LCC P166 (ebook) |
DDC 415.01/834–dc23/eng/20230313
LC record available at https://lccn.loc.gov/2022056326
LC ebook record available at https://lccn.loc.gov/2022056327
ISBN 978-1-107-13045-6 Hardback
Cambridge University Press & Assessment has no responsibility for the persistence
or accuracy of URLs for external or third-party internet websites referred to in this
publication and does not guarantee that any content on such websites is, or will
remain, accurate or appropriate.
Index 983
20.24 Derived tree for (10) before (top) and after (bottom) final
edge feature unification 766
20.25 RRG template schemas for non-subordinate nexus types 767
20.26 Two ways of compositionally deriving
cosubordination constructions 767
20.27 Core cosubordination via sister adjunction: analysis of (13) 768
20.28 Core cosubordination analysis of (14) derived by
wrapping substitution 769
20.29 Syntactic representations of the examples in (15) 769
20.30 Composition of templates by wrapping substitution for
(15a) and (15b) 770
20.31 Composition of the syntactic representation of (16a) 770
20.32 Subordination via simple and wrapping substitution for
(17a) and (17b), respectively 771
20.33 Derived tree for (18) 772
20.34 Derivation for (18) 773
20.35 Derivation for (19a) 774
20.36 Derivation for (21) 775
20.37 Syntactic tree for (24) 776
20.38 Derivation for (25a) 777
20.39 Linking syntax and semantics: derived tree 778
20.40 Anchoring the default transitive template with
‘smashed’ 779
20.41 Syntactic composition for (26) after argument linking 779
21.1 The Vauquois triangle (based on Dorr et al. 2006: 384) 794
21.2 MT–RRG-based interlingua bridge approach (from Nolan
and Salem 2011: 315) 795
21.3 The RRG interlingua for UniArab (from Nolan and Salem
2011: 315) 795
21.4 The conceptual architecture of the UniArab system (from
Nolan and Salem 2011: 321) 796
21.5 Dialogue management framework for a language
understanding system supporting conversational agents
and a dynamic common ground 800
21.6 Conceptual framework of the conversational software
agent (from Panesar 2017: 190) 801
21.7 The avatar model (from Murtagh 2019b: 95) 802
21.8 The hand articulator in 3D space with nodes on fingers
and thumb (from Murtagh 2019b: 98) 803
21.9 The architecture of FunGramKB (source: www.fungramkb
.com) 806
22.1 Verb structure in Cheyenne (modified from Corral
Esteban 2017: 310) 825
22.2 Layered structure of the clause with constituent and
operator projections 826
22.3 Representation of direct core arguments (cf. 28) 836
22.4 Representation of an applicative construction with a
derived two-place verb 837
TABLEAUS
7.1 Transitive constructions in Icelandic page 333
7.2 ‘Dative-subject’ constructions in Icelandic 333
7.3 Transitive constructions in Warlpiri 334
7.4 Case syncretism in Kabardian 353
7.5 Case syncretism in Old Persian 353
Beginning Advanced
Introduction: §1.1 x x
The structure of simple sentences:
Basic clause structure: §1.2–1.2.1 x x
Lexical and syntactic categories: §1.2.2 x (Ch. 2)
Operators: §1.2.3–1.2.3.2, 1.2.3.4 x (Ch. 2) x (Ch. 2)
Status of copular be: §1.2.3.3 x
Peripheries: §1.2.4 x x
Phrasal adjuncts: §1.2.4.1 x (Ch. 10) x (Ch. 10)
Non-phrasal adjuncts: §1.2.4.2 x (Ch. 9) x (Ch. 9)
Structure of phrases: §1.2.5 x (Ch. 10) x (Ch. 10)
Syntactic templates: §1.2.6 x x
Formalization x (Ch. 20)
Structure of words: §1.2.7.0 x x (Ch. 8)
Lexical integrity hypothesis: §1.2.7.1 x
Head-marking: §1.2.7.2 x
Information structure
Basic concepts: §1.5 x (Ch. 11) x (Ch. 11)
(cont.)
Beginning Advanced
Applying RRG
Historical linguistics x (Ch. 17)
Language acquisition x (Ch. 18)
Neurolinguistics x (Ch. 19)
Computational linguistics x (Ch. 21)
Grammatical sketches x (Ch. 22–26)
For Role and Reference Grammar (henceforth RRG), capturing the extent of
variation in the grammars of the world’s languages is as important a goal of
linguistic theory as identifying and explaining the properties that all gram-
mars share. The first work which aimed to provide a rigorous, comprehen-
sive and coherent analysis of the syntax of a number of typologically
dissimilar languages, introducing many of the constructs and principles
which would then be developed into tenets of RRG, was Foley and Van
Valin’s (1984) Functional Syntax and Universal Grammar. Almost four decades
after the publication of that volume, the Cambridge Handbook of Role and
Reference Grammar sets forth to fulfil a more ambitious set of objectives. The
volume is an up-to-date presentation of the framework, assuming very little
familiarity, if any, on the part of the reader, while also introducing the
many developments which RRG has undergone since the publication of the
following manuals and collections: Van Valin and LaPolla (1997), Van Valin
(2005), Pavey (2010) and Mairal Usón et al. (2012) (for further relevant work
see the bibliography available here: https://rrg.caset.buffalo.edu/). In pursu-
ing this objective, the volume is meant to be, quite literally, a handbook,
that is, the most complete current treatment of RRG and the first point of
reference for any researchers and teachers interested in this framework. Van
Valin’s Chapter 1 provides a broad overview of the principles and workings
of the framework, while the chapters in Parts II and III adduce more detailed
discussions of simple and complex sentences, respectively. Part II deals with
the lexicon, lexically motivated alternations, the structure of the word, and
semantic decomposition (Cortés-Rodríguez; Mairal Usón and Faber;
Peterson; Watters); semantic macroroles and grammatical relations
(Kailuweit; LaPolla); important facets of the linking (Latrouite and Van
Valin; Nakamura); adpositional, adverbial and mimetic constructions
It is hoped that these different audiences will find the handbook instructive
and stimulating, and that the descriptions and analyses presented here will
engender constructive scientific debate both within the RRG community
and further afield.
The volume is the result of long-term collaborative efforts of many scholars
worldwide. The editors would like to thank the reviewers of the handbook
proposal, and of the individual chapters, the many chapter authors, and last
but by no means least, the whole community of RRG linguists, who enthu-
siastically and unfailingly supported the project through its many stages and
vicissitudes, helping the editors to see it to its successful completion. At
Cambridge University Press, the editors thank Helen Barton, Isabel Collins
and Stephanie Taylor for their prompt, effective and friendly support.
also reflect the linear ordering requirements of the language and the whole
range of clausal layers that it is endowed with (for the non-universal ones,
which symmetrically occur to the left and the right of the Core, see
Chapter 1). These templates are one of the constructional features of the
framework, and we shall return to other such features below.
The scope of aspect, negation, tense, deontic and epistemic modality, etc.
over different layers of the clause is a language universal for RRG (see
Chapter 9 for similar considerations regarding the scope of adverbs and
ideophones). Indeed, Foley and Van Valin (1984: 208–224) (see also Van Valin
and LaPolla 1997: 46–51; Van Valin 2005: 12) make the strong empirical
claim that the linear order of operators in syntax follows from their scope.
This is the Natural Serialization Principle, which was supported by the results
of a typological survey reported in Bybee (1985). Thus, operators with
narrower scope (e.g. aspect) are predicted always to be closer to the nucleus
than operators with wider scope (e.g. deontic modality). Although similar
claims are made in Functional Syntax (Dik 1978, 1980, 1989), as well as in
Cartographic proposals (Cinque 1999), it is essential to note that the con-
stituent projection can only host predicative and referential units in RRG,
and, thus, it cannot accommodate any functional projections, comparable
to IP, or indeed Mood(. . .)P, Tense(. . .)P, Aspect(. . .)P, etc. Instead, the universal,
and strictly hierarchical, array of operators is represented in the operator
projection (see Chapter 1).
The fact that operators are represented separately from the clausal layers
upon which they have scope has brought to light an interesting, and hith-
erto unexplained, property of language change (Matasović 2008 and
Chapter 17). In accordance with the principle of unidirectionality, which
has been explored from many theoretical perspectives, operators of
narrower scope tend to grammaticalize as operators of broader scope (aspect
> tense; deontic modality > epistemic modality). In RRG terms, this means
that nuclear operators can only develop from nuclear operators, but can
themselves develop into nuclear, core or clause operators. In turn, core
operators can only develop from nuclear or core operators, but can them-
selves develop into core or clause operators. This is what Matasović (2008)
calls the centrifugal direction of grammaticalization. However, the gramma-
ticalization of verbal forms proceeds from lexical units to clitics and affixes,
thus moving in a direction which is, in effect, centripetal, and opposite to
that of the change in semantic scope. Thus, by disentangling operators from
the syntactic units upon which they have scope, work in RRG has uncovered
the opposite directions of the two unidirectional processes which character-
ize grammaticalization. This is an important contribution of the framework
to the study of language change.
The layers of clause structure, and the respective operator scope, play a
key role in the RRG theory of predicate and clause linkage (for which see
Chapters 13, 14, 15, 16). This theory relies on a distinction which, to our
knowledge, has no direct equivalent in other frameworks, namely that
accusative alignment privileges the high end of the hierarchy in syntax (see
above), ergative alignment privileges the low one. In split alignment, the
high end provides the privileged syntactic argument of some constructions,
though not others. The event structure hierarchy is one and the same in all
grammars, and, indeed, there is reason to consider it to be the linguistic
correlate of a cognitive universal. However, the study of argument realiza-
tion across languages indicates that both ends can be syntactically
unmarked in different languages or constructions, and this, in our view,
supports the parallel architecture approach. Interestingly, evidence from
neuroscience also corroborates the idea that the lexicon is an independent
module in linguistic competence (see Chapter 19). Therefore, the linking
will remain at the very centre of the RRG conception of grammar, and
indeed work in RRG has over the years sought to refine the system of lexical
decomposition which the event-structure hierarchy is a derivative of (Mairal
Usón and Faber 2002; González-Orta 2002; Van Valin and Mairal Usón 2014;
Bentley 2019, among others).
As for the role of constructions, these are templates which define the
unique syntactic, morphological, semantic and pragmatic features of indi-
vidual constructions in individual languages. These templates – called
‘Constructional Schemas’ – need not include any of the general principles
which are valid in the grammar of a given language, or across languages,
but rather are sets of specific instructions which, combined with the
general linking principles, constitute the grammar of a particular lan-
guage. Constructions thus play a different role in RRG than in
Construction Grammar (Goldberg 1995, 2006), and, crucially, they do not
in any way reduce the role of the linking or of its components, including
the lexicon.
In this section we have introduced key aspects of the formalism
developed by RRG to explore how different languages express linguistically
salient meaning and communicate it in context. We have shed light on
similarities and differences with other frameworks, and we have reflected
on the role of RRG in current debates in linguistic theory. We hope that
these observations will encourage researchers of various theoretical per-
suasions to read further.
References
Foley, William A. and Robert D. Van Valin, Jr. 1984. Functional Syntax and
Universal Grammar. Cambridge: Cambridge University Press.
Givón, Talmy. 1980. The binding hierarchy and the typology of comple-
ments. Studies in Language 4: 333–377.
Goldberg, Adele E. 1995. Constructions: A Construction Grammar Approach to
Argument Structure. Chicago: University of Chicago Press.
Goldberg, Adele E. 2006. Constructions at Work. Oxford: Oxford University
Press.
González-Orta, Marta Maria. 2002. Lexical templates and syntactic variation:
The syntax-semantics interface of the Old English speech verb secgan. In
Mairal Usón and Pérez Quintero (eds.), 281–302.
Gruber, Jeffrey S. (1965) Studies in Lexical Relations. PhD dissertation, MIT,
Cambridge, MA.
Jackendoff, Ray. 1976. Toward an explanatory semantic representation.
Linguistic Inquiry 7(1): 89–150.
Jackendoff, Ray. 2002. The Architecture of the Language Faculty. Cambridge, MA:
MIT Press.
Kamp, Hans and Uwe Reyle. 1993. From Discourse to Logic. Hingham, MA:
Kluwer.
Lambrecht, Knud. 1994. Information Structure and Sentence Form: Topic, Focus,
and the Mental Representation of Discourse Referents. Cambridge: Cambridge
University Press.
Levin, Beth and Malka Rappaport Hovav. 2005. Argument Realization.
Cambridge: Cambridge University Press.
Mairal Usón, Ricardo and Pamela Faber. 2002. Functional Grammar and
lexical templates. In Mairal Usón and Pérez Quintero (eds.), 39–94.
Mairal Usón, Ricardo and María Jesus Pérez Quintero (eds.). 2002. New
Perspectives on Argument Structure in Functional Grammar. Berlin: Mouton de
Gruyter.
Mairal Usón, Ricardo, Lilián Guerrero and Carlos González (eds.). 2012. El
funcionalismo en la teoría lingüística. La Gramática del Papel y la Referencia.
Introducción, avances y aplicaciones. Madrid: Akal.
Martín Arista, Javier. 2009. A typology of morphological constructions. In
Christopher Butler and Javier Martín Arista (eds.), Deconstructing
Constructions, 85–116. Amsterdam: John Benjamins.
Matasović, Ranko. 2008. Patterns of grammaticalization and the layered
structure of the clause. In Rolf Kailuweit et al. (eds.), New Applications of
Role and Reference Grammar, 45–57. Newcastle upon Tyne: Cambridge
Scholars Publishing.
Pavey, Emma L. 2010. The Structure of Language. An Introduction to Grammatical
Analysis. Cambridge: Cambridge University Press.
Perlmutter, David M. (ed.). 1983. Studies in Relational Grammar 1. Chicago:
University of Chicago Press.
Perlmutter, David M. and Brian D. Joseph (eds.). 1990. Studies in Relational
Grammar 3. Chicago: University of Chicago Press.
Abbreviations
1.1 Introduction
* This research was supported in part by a Fellowship from the Max Planck Society and from the German Research
Foundation via CRC 991 ‘The structure of representations in language, cognition and science.’ I would like to thank Delia
Bentley, Jürgen Bohnemeyer, Fabricio Gerardi, Ranko Matasović, Wataru Nakamura, Rainer Osswald, Mitsuaki Shimojo, and
Jan Ullrich for comments on an earlier draft.
(1) [PROPOSITION [PREDICATE X ][ARGUMENT (Y1), (Y2), . . .][ADJUNCT (Z1), (Z2), . . .]]
These semantic units underlie the central syntactic units of the layered
structure: the notion of predicate motivates the nucleus, which houses
the predicating element (the predicator), the core, which consists of the
nucleus plus the arguments of the predicating element, and the peripher-
ies, which house the adjuncts. This can be represented as in (2) and applied
to an English example in (3).
(2) [CLAUSE [CORE (Y1) [NUCLEUS X ] (Y2), . . .][PERIPHERYcore (Z1)] CORE][PERIPHERYclause (Z2)]
CLAUSE]
(3) [CLAUSE [CORE Mary [NUCLEUS ate ] a cookie ][PERIPHERYcore after lunch] CORE]
[PERIPHERYclause despite her diet.] CLAUSE]
Nucleus, core and periphery are syntactic units which are relatively motiv-
ated by the semantic constructs in (1) but are not purely semantic. While in
the prototypical case, the nucleus and the predicating element (predicator)
are co-extensive, it is possible to have non-predicators in a nucleus. A classic
example of this is noun incorporation, in which one of the arguments of the
predicator occurs in the nucleus along with the predicator itself, for
example Lakhota [čháŋ kiŋ [NUC kaksá]] (wood the chop) ‘chop the wood’ vs.
[[NUC čhaŋkáksa]] ‘chop wood’, ‘do wood-chopping’. In the non-incorporated
form the object argument čháŋ kiŋ ‘the wood’ occurs as an independent
argument, whereas in the incorporated form the noun čháŋ ‘wood’ occurs
without an article and is compounded with the predicator kaksá ‘chop’,
yielding čhaŋkáksa. Multisyllabic words in Lakhota tend strongly to have
stress on the second syllable, as with kaksá ‘chop’, and in the incorporated
form the stress has shifted to the second syllable, which is to be expected
due to the incorporation of čháŋ ‘wood’ (which loses its own accent), which
creates a trisyllabic word.
With respect to the core, in the prototypical case, a semantic argument of
the predicator in the nucleus occurs in it, and an XP in the core is a semantic
argument of the predicator. However, there are cases in which a semantic
argument does not occur in the core, for example the agent in a passive
construction appears in the core-level periphery, or a wh-expression may
occur in a position outside of the core in some languages (e.g. Figure 1.4).
Conversely, there are XPs occurring in the core which are not semantic
arguments of the predicator in the nucleus. Examples include dummy
arguments, like those that occur with meteorological verbs in some lan-
guages (e.g. It is raining, Il pleut (French), Es regnet (German)), and ‘raised’
arguments in ‘raising constructions’ (e.g. Mary seems to like okra, in which
Mary occurs as a syntactic argument of the core headed by seems but is a
semantic argument of like).
Peripheries canonically contain adjuncts which modify a specific layer of
the clause. In (3), for example, the temporal adjunct after lunch modifies the
core Mary ate a cookie and occurs in the core-level periphery, and the conces-
sive adjunct despite her diet modifies the clause Mary ate a cookie after lunch and
occurs in the clause-level periphery. There is at least one case, however,
where the XP in a periphery is not an adjunct modifier but rather a semantic
argument of the predicator in the nucleus, namely, the agent in a passive
construction. Moreover, there are constructions in which the adjunct repre-
sents a participant in the state of affairs which is not a semantic argument
of the nucleus, for example the deputative beneficiary marked by for in Bill
went to the store for Mary.
The distinctions in (2) are better represented in a tree diagram, as in
Figure 1.2.
The nucleus contains the predicator and may be modified by an adverb in
the nuclear-level periphery.1 The core minimally contains the nucleus (and
its periphery, if present), which would be the case for predicators which
have no arguments (e.g. Lakhota maǧážu ‘it is raining’), but usually there are
one or more argument expressions. It may optionally be modified by adverbs
or adjunct adpositional phrases in the core-level periphery. The clause
contains the core (plus its periphery, if present), and, like the nucleus and
core, can be modified optionally by peripheral adverbs or adpositional
phrases. The sentence node reflects the traditional idea that a sentence
may consist of multiple clauses; its function will be clarified below. Linear
order is not relevant to the definition of the units of the LSC; the order of
elements in Figure 1.2 was chosen for clarity and is in no way privileged.
This figure presents the universal features of the LSC.
The structure of (3) is given in Figure 1.3. Eat is the predicator in the
nucleus, Mary and a cookie are the core arguments, and the two PPs, after
lunch and despite her diet, occur in peripheries. There is no VP in the structure,
despite the well-known evidence that some constructions in English clearly
involve a ‘VP’-like grouping, for example I expected to find someone washing the
dishes, and [‘VP’ washing the dishes] was Bill. This is because ‘VPs’ are not
universal; some languages have constructions with them and others do
(4) a. A: What do you think of the new novels by Suzanne Collins and John
Grisham?
B: Collins’ book I really enjoyed, but Grisham’s I found boring.
b. A: Have you read anything good lately?
B: Suzanne Collins’ new book I really liked, but John Grisham’s latest
offering I didn’t care for.
In English the subject is the first (non-oblique) argument in the core, and
so a non-wh XP appearing immediately before it is outside of the core and is
therefore in the PrCS. XPs in the PrCS, regardless of whether they are wh-
expressions or not, are not set off by an intonation break, and they never co-
occur with a resumptive pronoun in languages like English.4
The existence of the PrCS in languages with nucleus-initial default word
order (e.g. Tagalog (Austronesian, Schachter and Otanes 1972); Tzotzil
(Mayan, Aissen 1987); and Wari’ (Chapakuran, Brazil, D. Everett 2008)) is
straightforward to diagnose: an XP before the nucleus without an inton-
ation break or an independent resumptive pronoun is in the PrCS. This is
illustrated by the Tagalog examples in (5) from Schachter and Otanes (1972:
496–498).
The default word order in Tagalog is given in (5a), in which the nucleus
containing the predicator tumawa ‘laughed’ occurs initially in the clause.
It is followed by siya ‘3sg.nom’, which is a clitic pronoun that must
occur in second position in the clause; in (5a) second position is imme-
diately following the nucleus. In (5b) the temporal adverb kahapon ‘yes-
terday’ occurs before the nucleus, and it is followed by siya, the second
position clitic. Based on clitic placement and a number of other criteria,
Latrouite and Van Valin (2021) show that the adverb is in the PrCS in
(5b). The same considerations lead to the conclusion that sa kaniya ‘at
him’ in (5b′) is likewise in the PrCS. It should be noted that the PrCS
element has a contrastive narrow focus reading like the English
example in (4b).
Languages with nucleus-final or extremely flexible word order present
something of a challenge with respect to establishing the existence of a
PrCS, because there is no fixed reference point for determining the onset of
the core, unlike in Tagalog or English. The best evidence is the obligatory
occurrence of wh-expressions at the beginning of the clause in wh-questions.
It is well known, however, that nucleus-final languages rarely have obliga-
tory displacement of wh-expressions to the beginning of the clause; they
either have only wh-in situ or optional displacement. The latter situation,
namely optional displacement to the beginning of the clause or in situ, is in
principle distinguishable from ‘free’ word or phrase order, since in the latter
situation a wh-expression could occur in any position in the clause; in
In Figure 1.4, there is an XP (as for Mary) before the wh-expression, and it
has quite different properties from XPs in the PrCS. It is normally set off by
an intonation break, represented orthographically by a comma, and if the
XP is an argument of the predicator in the nucleus of the clause, as in this
sentence, there is obligatorily a resumptive pronoun in the core of the
clause, in this case she. This position is termed the Pre-detached position
(PrDP). It is usually termed ‘the left-detached position’ in the linguistic
literature, including previous work in RRG. However, there is no left or
right in spoken language, only before and after. Only in written language
can one speak of left and right, and the label ‘left-detached position’ reflects
the fact that European languages are written from left to right. For scholars
writing in languages like Arabic and Farsi, on the other hand, it is the ‘right-
detached position’. There is no such problem with PrCS and PoCS, since ‘pre-’
and ‘post-’ refer to order in speech, not order on the printed page. Accord-
ingly, the problematic terms ‘left-detached position’ and ‘right-detached
position’ are replaced by PrDP and Post-detached position (PoDP), in
order to avoid confusion.6
The two detached positions are outside of the clause and are direct
daughters of the sentence node. Both are normally set off by an intonation
break, and both require a resumptive pronoun if the XP in the detached
position is an argument of the predicator in the nucleus of the clause.
Tagalog presents clear evidence that the PrDP is outside of the clause, as
exemplified in (7) from Schachter and Otanes (1972: 489).
The topic expression, si May, is set off by an intonation break and option-
ally marked by the topic marker ay (Nuhn 2021; Latrouite and Van Valin
2021). It must be in the PrDP, because it is followed by the adjunct wh-
expression kailan ‘when’, which is in turn followed by two second-position
clitics, the interrogative marker ba and the optional resumptive pronoun
siya ‘3sg.nom’. The placement of the clitics shows that kailan ‘when’ must be
clause-initial, namely in the PrCS, as in (5b,b′), and this means that si May is
outside of the clause and therefore in the PrDP. The structure of (7) is given
in (8).
(8) [S [PrDP Si May (ay)], [CL [PrCS kailan ]¼ba¼(siya) [CORE [NUC babalik] dito?] CL] S]
The constituents of the LSC, both obligatory and optional, minus the per-
ipheries, are given in Figure 1.5.
These constituents can be viewed in terms of whether they are universal
or non-universal. It has been argued that the nucleus, core and clause
‘spine’, plus peripheries, is semantically motivated and is based ultimately
on the necessary properties of language as a system of human communi-
cation. All languages make use of the nucleus, core and clause spine. On the
other hand, not all languages have detached positions or special positions
outside of the core and inside the clause. They are pragmatically motivated
rather than semantically motivated. While it would be impossible to form
propositions without predicators and arguments, it is possible to communi-
cate effectively without these special positions. Lakhota, for example, has a
PrDP but no PrCS or PoCS. Dyirbal, with its extremely (grammatically) free
word order, has no fixed positions with reference to which the PrDP or PrCS
could be defined; moreover, it lacks markers indicating topics and foci akin
to Japanese wa and ga, for example, which could be associated with these
distinctions (see Shimojo 2011). The packaging of information in terms of
special fixed positions for topics and foci is a useful but not an indispensable
feature of human language.
(9) a. Max is a very good lawyer. Predicator ¼ a very good lawyer [RP]
b. Max is (extremely) tall. Predicator ¼ (extremely) tall [ADJP]
c. Max is in the house Predicator ¼ in the house [PP]
In (10a) the predicator in the nucleus is a verb, and the single core
argument is a noun, suggesting that it is correctly analysed as an NP. In
(10b), on the other hand, the predicator is a noun and the head of the
argument expression is the verb tumawa ‘laughed’. Importantly, there is no
change in the form of the words in the two sentences: bata ‘child’ is an
argument in (10a) and the predicator in the nucleus in (10b), and tumawa
‘laughed’ is the predicator in the nucleus in (10a) and the head of an
argument expression in (10b). This contrasts with the English translation
of (10b), in which the verb-based argument expression is expressed as a
relative clause and the nucleus requires the auxiliary be due to the non-
verbal predicator. Thus, unlike in English, a verb can head a referring
expression without modification, and a noun can function as the predicator
without modification. The same pattern holds in the Lakhota sentences in
(11); both predicators take the same inflectional morphology, regardless of
the lexical category of the head, and the head of the ‘NP’ wačhí kiŋ hé ‘that
dancer’ is the verb wačhí ‘dance’.
These examples are no problem for the category-neutral notion of
nucleus, and they show that a category-neutral syntactic category for
referring expressions is needed as well. Van Valin (2005: 28) made an initial
suggestion which was further developed into the notion of RP in Van Valin
(2008).9 It might be objected that there is no need for this notion, because
these examples can be handled in terms of the already existing notion of
DetP. The argument expressions in (10) and (11) would be DetPs, not NPs,
and the complement to the determiner head of the phrase could be of
different categories. However, DetP is a problematic construct, given RRG’s
assumptions about syntactic structure. In order to have a DetP there must
be a determiner functioning as the head of the DetP, and in many lan-
guages there are minimal determiner systems; Russian, Mandarin and
Japanese, for example, have demonstratives but no articles. In such lan-
guages many sentences have arguments which lack determiners, and
therefore to justify calling them DetPs it would be necessary to posit
phonologically null determiners as heads of the DetPs. This is incompat-
ible with the RRG prohibition against phonologically null elements in the
syntax. RP, on the other hand, requires no such postulation of null deter-
miners. Thus in (10) and (11) the arguments are RPs, regardless of whether
the head is a noun or a verb.
Modifiers, in particular those modifying within RPs, do not necessarily
follow an endocentric pattern. The following examples are from D. Everett
(2008).
(13) a. [CLAUSE [RP Wíŋyaŋ kiŋ hená] [NUC wačhí-pi-kte-šni] CLAUSE] Lakhota
woman the those dance-pl-pot-neg
‘Those women will not dance.’
a′. [CLAUSE [RP *Wíŋyaŋ-pi kiŋ hená] [NUC wačhí-pi-kte-šni] CLAUSE]
woman-pl the those dance-pl-pot-neg
‘Those women will not dance.’
b. [CLAUSE [RP Wačhí-pi kiŋ hená] [NUC wíŋyaŋ-pi-kte-šni ] CLAUSE]
dance-pl the those woman-pl-pot-neg
‘Those dancers will not be women.’
b′. [CLAUSE [RP *Wačhí kiŋ hená] [NUC wíŋyaŋ-pi-kte-šni ] CLAUSE]
dance the that woman-pl-pot-neg
‘Those dancers will not be women.’
1.2.3 Operators
In Figure 1.4, which represents the structure of As for Mary, what did she eat
after lunch?, the auxiliary verb did is not attached to the LSC tree in any way.
The reason for this is that the LSC trees introduced so far contain only
lexical words and morphemes. Closed-class grammatical morphemes indi-
cating functional categories like aspect, tense, modality and mood (illocu-
tionary force) are not represented. These notions are termed operators in
RRG, and they modify specific layers of the LSC. The operators discussed in
RRG are given in Table 1.1.
A detailed discussion of them can be found in Peterson (Chapter 2 of this
volume), and the emphasis in this section will be on selected cases of the
interaction of operators and the LSC.
The most commonly encountered operators cross-linguistically are aspect,
deontic modality, negation, tense, and illocutionary force (IF).14 Following a
suggestion in Johnson (1987), they are represented in a separate projection
of the clause (the operator projection) which is more or less a mirror
image of the lexical tree, which is normally referred to as the constituent
projection. A revised version of the tree in Figure 1.4 with the operator
projection added is given in Figure 1.6.
Nuclear operators
Aspect
Negation
Directionals (only those modifying the orientation of an action or event without reference to
participants)
Core operators
Directionals (only those expressing the orientation or motion of one participant with reference to
another participant or to the speaker)
Event quantification
Modality (root modals, e.g. ability, permission, obligation)
Internal (narrow scope) negation
Clausal operators
Status (epistemic modals, external negation)
Tense
Evidentials
Illocutionary force (IF)
b. Wačhí-šni yo/ye!
dance-neg imp.Male/imp.Female
‘Don’t dance!’
Lakhota IF particles index the sex of the speaker (Ullrich 2011: 820–822):
those ending in -o are for male speakers, while those ending in -e are for
female speakers. He in (15a) was originally a female form but has come to be
used by both sexes; the male form is huwó. The base form for the assertion
marker for men is yeló, while women use ye; kštó is used by both sexes but
primarily by women (Ullrich 2011).
These morphemes raise an important issue with respect to the concreteness
restriction on syntactic representations in RRG. The interrogative and impera-
tive particles are obligatory in questions and commands, because without
them the utterance would not be interpreted as having those IF values. The
assertion particles are different; if they are not used, as in (11a), the utterance
is still interpreted as an assertion due to the lack of the interrogative and
imperative markers. Hence there are two ways to indicate an assertion in
Lakhota: use of an overt assertion marker, or use of no IF particle of any kind.
The first is unproblematic: there is an overt form, kštó for example, which can
be linked to the operator projection as the IF operator with scope over the
clause layer. The problem arises in the second case in which there is no IF
marking of any kind, yet the sentence has a definite IF value. It is necessary to
represent this, especially in the analysis of complex sentences; this will be
discussed in Section 1.3. This is illustrated by (16), the structure of which is
given in Figure 1.7; only the operator projections are represented.
The issue with the second representation is whether it violates the prohib-
ition against phonologically null elements in the syntax. There are good
grounds for concluding it does not. First, the proscribed null elements are
lexical categories or projections thereof; for example, null pronominals or
verbs are considered to be members of the lexical categories of pronominal
and verb.
The zero morpheme in (16) is a closed-class grammatical (function) mor-
pheme and is not a free form which can be manipulated in the syntax,
Figure 1.8 Examples of the ordering of aspect and tense markers in different languages
because no ordering relation between the two can be established; hence the
ordering constraint does not apply in this case.
This constraint applies to the ordering of operators modifying different
layers of the LSC, but it does not restrict the order of multiple operator
morphemes in a given layer. At the nuclear and core layers the possibilities
are limited, namely, aspect and negation at the nuclear layer, and negation
and deontic modality at the core level. With respect to aspect and negation,
it would seem that the only logical possibility is negation > aspect; in a
system with perfective and imperfective aspect, NEG [V-(im)perfective] is
meaningful (i.e. ‘not [(in)complete state of affairs]’, e.g. ‘not [finished
raining]’), but the other possibility is not (i.e. ‘(in)complete [NEG state of
affairs]’, e.g. ‘finished [not raining]’). Van Valin and LaPolla (1997: 46) give an
example of the interaction of aspect and negation from Qiang, a Tibeto-
Burman language, and as predicted, negation has scope over aspect, and the
order is negation-aspect-V.
At the core level, on the other hand, both scopes are possible. In a sentence
like John must not leave the door open, the deontic modal can have scope over
negation (i.e. ‘John is obliged not to leave the door open’), or the deontic modal
can be in the scope of negation (i.e. ‘John is not obliged to leave the door open’).
Interestingly, the second reading is stronger if the negation is contracted, as in
John mustn’t leave the door open. This variation in scope is not reflected in the
order of operator morphemes in a simple sentence, unlike in the paraphrase
which contains two cores (see Section 1.3). It is, rather, signalled prosodically.
The two readings can be distinguished clearly by different prosodic chunking
of the sentence: {John must}{not leave the door open} vs. {John must not}{leave the
door open}. The negative morpheme in English makes up for its fixed position
by being able to scope in either direction.
Multiple operators regularly co-occur at the clause level, as in Figure 1.6
in which tense and IF co-occur. There are four clausal operators (IF,
evidentials, status (clausal negation, epistemic modality) and tense), and
they fall into two groups: tense and status, on the one hand, and eviden-
tials and IF, on the other. Tense and status are propositional modifiers and
occur in both main and subordinate clauses. Evidentials and IF are utter-
ance modifiers and are restricted to main clauses. In agglutinative lan-
guages in which operator morphemes occur in sequence, as in Lakhota,
Kewa, Turkish, Tiwi and Qiang, the IF morpheme occurs farthest from the
nucleus, if it has a fixed position; this is illustrated in the Lakhota
examples in (15) and (16). In terms of the LSC operator projection, the IF
and evidential operators modify the clause node which is immediately
dominated by the sentence node; there is no such restriction on tense or
status, hence they can occur freely in subordinate clauses. There is no
scope asymmetry between tense and status; they modify at the same level,
but because the tense and status morphemes must be linearized, lan-
guages show variation in the order of these morphemes.
(19) a. [S [CL [PrCS What ] {did} [CORE Sally [NUC borrow ] from Bill? CORE] CL] S]
b. [S [PrDP *As for whati ][CL{did} [CORE Sally [NUC borrow ] iti from Bill? CORE] CL] S]
This felicitous pair of utterances shows that the RP a new Tesla can be the
focus of the assertion and therefore denied. This is not possible with an RP in
the PrDP, as in (21a), but is fine with an XP in its default position, as in (21b),
or in the PrCS, as in (21c).
(21) a. Speaker 1: On Thursday, Mary will meet the journalist at the library.
Speaker 2: No, at Starbucks/No, the social worker/No, Bill will/??No, Friday
b. Speaker 1: Mary will meet the journalist at the library on Thursday.
Speaker 2: No, at Starbucks/No, the social worker/No, Bill will/No, Friday
c. Speaker 1: Thursday Mary will meet the journalist at the library.
Speaker 2: No, at Starbucks/No, the social worker/No, Bill will/No, Friday
(22) a. Speaker 1: On Thursday, will Mary meet the journalist at the library?
Speaker 2: No, at Starbucks/No, the social worker/No, Bill will/??No, Friday
b. Speaker 1: Will Mary meet the journalist at the library on Thursday.
Speaker 2: No, at Starbucks/No, the social worker/No, Bill will/No, Friday
c. Speaker 1: Thursday will Mary meet the journalist at the library.
Speaker 2: No, at Starbucks/No, the social worker/No, Bill will/No, Friday
Thus, the features distinguishing the PrDP and the PrCS are directly related
to the fact that the scope of the IF operator is the clause. The same holds for
the PoDP and PoCS; as the Dhivehi examples in (6) show, the PoCS can host wh-
expressions analogous to (19a) and contrastive focus phrases analogous to (4b).
The PoDP, like the PrDP, cannot be asserted or questioned, as illustrated in
(23a, b), nor can it host wh-expressions, as shown in (24).
(23) a. Speaker 1: Mary bought them ice cream cones, the kids.
Speaker 2: No, some popcorn/No, Bill did/??No, the adults.
b. Speaker 1: Did Mary buy them ice cream cones, the kids?
Speaker 2: No, some popcorn/No, Bill did/??No, the adults.
In (25a), the tense operator is realized on the passive auxiliary be, and
its core-internal linear position marks declarative IF. In this construction
was is attached to both projections: the tense and IF operators connect it
to the operator projection, while the be to which these operators are
attached is structurally required for passive nucleus formation.20 In
(25b), the auxiliary was is attached to the operator projection three times:
(1) it is part of the two-part progressive aspect morpheme, beþ-ing; (2) it
signals past tense; and (3) its core-internal position in the linear string
indicates declarative IF. Passive be is not connected to the operator pro-
jection; it hosts -ing, which is part of the expression of progressive aspect
and is attached to the operator projection. Thus, passive be is always part
of the constituent projection and may also be connected to the operator
projection, whereas progressive be can only be connected to the operator
projection.
This follows from the fact that progressive be occurs only when the
progressive aspect is expressed. Passive be, like copular be, is structurally
required for proper nucleus formation and occurs even when there are no
operators, as in the infinitives in (26).
It should be noted that there are other auxiliary verbs which can fulfil be’s
structural role while contributing semantically to the predication. For
clauses with adjectival predicates, seem and appear can replace be as in Bill
is/seems/appears happy/stronger/healthy. Become and get can also serve this func-
tion: Mary became despondent, Sally got stronger. Get can also replace be in a
passive construction, as in Max got interviewed by CNN during the insurrection.
The scope of negation is tied to the focus structure of the clause, and
consequently the interpretation depends on the context in which the
sentence occurs.
1.2.4 Peripheries
In the earlier versions of the LSC there was only a single periphery, which
was attached at the core level. This proved to be wholly inadequate, and in
Van Valin (2005) it was proposed that each of the layers of the spine of the
LSC was modified by a periphery containing modifiers appropriate to the
layer. Moreover, a distinction was made between phrasal and non-phrasal
adjuncts, phrasal adjuncts being PPs and non-phrasal adjuncts being
adverbs, in simple sentences.21 In complex sentences, phrasal adjuncts also
include what are traditionally called adverbial clauses (to be discussed in
Section 1.3).
The adjunct PP tosyokan de ‘in the library’ is linearly within the core, and the
LSC diagram necessarily involves crossing tree branches, as in Figure 1.10.
The PP tosyokan de ‘in the library’ is in the core-level periphery, because it
expresses the location of the event described by the core, that is, Hanako’s
reading a book, and consequently the representation of the hierarchical
structure and the linear sequence of words leads to the crossing branches.
Such crossing branches are allowed in RRG, because the tree represents
semantic dependencies primarily rather than purely syntactic constituent
structure.
German presents an interesting case, as there has long been a controversy
as to whether it is fundamentally verb-medial or verb-final. With respect to
the default placement of adjunct PPs, it patterns with Japanese and not with
English.
(30) a. Max hat nach der Auseinandersetzung mit Maria eine ganze Flasche Wein getrunken.
has after the argument with a entire bottle wine drunk
‘Max drank an entire bottle of wine after the argument with Mary.’
b. Max trank nach der Auseinandersetzung mit Maria eine ganze Flasche Wein.
drank after the argument with a entire bottle wine
‘Max drank an entire bottle of wine after the argument with Mary.’
In (30a) the nucleus is core- (and clause-) final, as in Japanese, and the order
of arguments and adjuncts is the same, too: subject, core-level adjunct PP,
direct object. In (30b) the nucleus is in second position in the core, as in
English, but the default order of arguments and adjuncts is the same as in
(30a) and Japanese. On the other hand, the word order in (30b) is impossible
in English; nothing is allowed to occur between the nucleus and the
following direct core argument (‘direct object’).22
Phrasal adjuncts may also appear in positions other than their default
position. In English and many other languages, they may appear in the PrDP
and function as a frame-setting topic.
(31) a. After the party, Sam went home and watched TV for a while.
a′. After the party, where did Sam go?
b. Because of the cold weather, vaccine distribution has been slowed down
significantly.
b′. Because of the cold weather, has vaccine distribution been adversely affected?
The adjunct PP in the PrDP does not express the topic of the sentence, unlike
in Figure 1.4 or (7), because (31a) is not about the party but rather about Sam
and (31b) is not about the cold weather. Rather, after the party and because of
the cold weather set the frame of reference in terms of which the following
(32) a. At the mall Sally couldn’t find what she wanted, but in a small boutique
she found it.
b. Before lunch Bill drank only coffee, but after lunch he ate a lot of snacks.
These initial PPs are not set off prosodically like the ones in (31) and are part
of the assertion, that is, they are within the scope of the IF operator, and are
therefore in the PrCS. This entails that they do not occur in question,
because the PP is in the actual focus domain (AFD; see Chapter 11).
(33) a. ??Before lunch did Bill drink only coffee, but after lunch did he eat a lot
of snacks?
b. *Before lunch what did Bill drink, but after lunch what did he eat?
Both examples are fine if there is a pause after the initial PPs, which
indicates that the structure is the same as in (31), not (32).
It was mentioned with respect to (3) in Figure 1.3 that the order of adjunct
PPs is constrained by the layer they modify, with core-level modifiers preced-
ing clause-level modifiers in a right-branching language like English. When
there are multiple adjunct PPs which modify at the same level, however, the
order is ‘free’ and influenced by information structure, as in (35).
(34) a. Mary ate a cookie after lunch despite her diet. [¼(3)]
a′. ??Mary ate a cookie despite her diet after lunch.
b. Sally met Max in the library because of the cold weather.
b′. ??Sally met Max because of the cold weather in the library.
(36) a. Vaccine distribution has been adversely affected despite the government’s
best efforts because of the cold weather.
b. Vaccine distribution has been adversely affected because of the cold
weather despite the government’s best efforts.
PPs headed by after and in are core-level modifiers, while those headed by
despite and because of are clause-level modifiers, and consequently the
ordering restriction in (34) and the lack thereof in (35) and (36) follows from
the analysis of phrasal adjuncts in terms of the LSC.
(37) a. Mary quietly ate a cookie after lunch in the garden despite her diet.
b. Mary ate a cookie quietly after lunch in the garden despite her diet.
c. Mary ate a cookie after lunch quietly in the garden despite her diet.
d. Mary ate a cookie after lunch in the garden quietly despite her diet.
e. ??Mary ate a cookie after lunch in the garden despite her diet quietly.
f. Quietly(,) Mary ate a cookie after lunch in the garden despite her diet.
(38) a. Mary evidently ate a cookie after lunch in the garden despite her diet.
b. ?Mary ate a cookie evidently after lunch in the garden despite her diet.
c. ?Mary ate a cookie after lunch evidently in the garden despite her diet.
d. Mary ate a cookie after lunch in the garden evidently despite her diet.
e. Mary ate a cookie after lunch in the garden despite her diet evidently.
f. Evidently, Mary ate a cookie after lunch in the garden despite her diet.
The oddity of (38b) and (38c) results from having a clause-level modifier
occurring in the midst of core-level modifiers. There is a striking contrast in
acceptability between (37e) and (38e), which is a function of the difference in
the level of modification (core vs. clausal) of the adverbs. The two ordering
possibilities in (38d, e) show that it makes no difference with respect to the
sequence whether the adjuncts are phrasal or not, as long as they are
modifiers of the same layer.23
As (37) and (38) make clear, adverbs can occur in positions which are not
part of the default periphery for phrasal adjuncts, and it was proposed in
Van Valin (2005) that all adjuncts not in the default periphery, the Pr/PoDP
or the Pr/PoCS are in a periphery specific to them. This is exemplified in
Figure 1.11.
The LSC is crucial for the analysis of the scope of modification of both
phrasal and non-phrasal adjuncts, but it has significant limitations, namely,
there are only three layers but there are many more types of adverbs. This is
illustrated in (39); the list of types of adverbs is not exhaustive.
Two points deserve mention. First, a given adverb can be used in more than
one category, as in the well-known example of cleverly as in Mary hid the
jewellery cleverly (manner-performance) vs. Cleverly, Mary hid the jewellery
(speaker attitude/judgement) vs. Mary cleverly hid the jewellery (ambiguous).
Second, even though the semantic domains of operators and adverbs are
similar, adverbs are not operators; adverbs are open-class lexical items,
whereas operators are closed-class grammatical elements. This is particu-
larly important when it comes to distinguishing the operator tense and
temporal adverbs; the former is a clausal operator, whereas the latter is a
core-level modifier. This can be seen clearly in sentences like (40).
(40) [CLAUSE [CORE1 Fred told Sally][ yesterday] [CORE2 to meet him][ in Paris Friday]].
This sentence consists of a single clause with a single tense operator which is
made up of two cores, each of which is modified by a different temporal
adverb. Each temporal adverb modifies the core it occurs with, while the
tense operator has scope over the entire clause.
It is reasonable to speculate that the largest class of adverbs is manner
adverbs, which has a number of subtypes, all of which are core-level modi-
fiers. Despite them all being core modifiers, the ordering among them does
not appear to be unconstrained.
(41) a. Mary will intentionally speak more clearly tomorrow.
b. ?Mary will tomorrow speak more clearly intentionally.
c. Mary will speak more clearly intentionally tomorrow.
d. Mary will speak intentionally more clearly tomorrow.
e. *Mary will more clearly speak intentionally tomorrow.
f. ?Mary will intentionally speak tomorrow more clearly.
g. ?Mary will tomorrow speak intentionally more clearly.
There are two manner adverbs, intentionally (intent) and clearly (perform-
ance), and one temporal adverb, tomorrow, in these sentences. Clearly seems
In (44) there is an adjunct PP, per sbaglio ‘by mistake’, which appears
between the PrCS RP il libro ‘the book’ and the core-initial subject Luca.
Normally, phrasal adjuncts occur after the core, but in this case, the adjunct
PP is in its own periphery, analogous to the non-phrasal adjunct in
Figure 1.11.
The Tagalog example is from Schachter and Otanes (1972: 498–499); it
involves a construction they label ‘non-emphatic inversion’, in which an
adverbial is displaced from its normal post-core position to the beginning of
the clause.
In the discussion of (5) from Tagalog, it was pointed out that the personal
pronouns are second-position clitics, and in (45b) nila ‘3plgen’ follows bigla
‘suddenly’, which shows it is in clause-initial position. It might be concluded
that it is in the PrCS, but it lacks the contrastiveness of PrCS XPs in Tagalog.
In fact, (46) shows that it is in the same place as per sbaglio in (44): between
the PrCS and the core.
The default position for the two adjuncts is given in (46a); they both follow
the single core argument. In (46b), on the other hand, they appear before the
nucleus, with the temporal adjunct kanina-ng umaga ‘this morning’ in
clause-initial position in the PrCS, as indicated by the contrastive interpret-
ation (cf. (5b)), and the adverbial mabilis ‘quickly’ between it and the core-
initial nucleus nagtrabaho ‘worked’. A comparison with (44b) reveals that the
structures are completely parallel.
It was mentioned earlier that in German the default location for adjuncts,
both phrasal and non-phrasal, is after the subject and finite verb or auxiliary
and before the remaining elements in the core. One of the relatively unique
features of German main clause structure is what is traditionally character-
ized as the ‘prefield’, the clause-initial position before the finite verb or
auxiliary which must be obligatorily filled. Wh-expressions normally occur
in it in questions, suggesting it is the PrCS, but in many sentences, the XP in
it is not contrastive or information-structurally special in any way. In (30),
for example, Max is a normal, topical subject in the prefield. In (47) it is
occupied by an adjunct which need not have any special information-
structural status.
(47) a. Nach der Vorlesung hat Dieter in der Bibliothek Almuth getroffen.
after the lecture has in the library met
‘After the lecture Dieter met Almuth in the library.’
b. Gestern hat Dieter in der Bibliothek Almuth getroffen.
yesterday has in the library met
‘Yesterday Dieter met Almuth in the library.’
Ruhnau (2011) gives an RRG analysis of the prefield and argues that in
(47a, b) the clause-initial adjunct is a peripheral constituent, not in the
PrCS. The prefield is thus rather heterogeneous structurally; it can be a
core-argument RP, as in (30), or PP, it can be the PrCS in wh-questions and
contrastive topic or focus sentences, or it can be an adjunct in
a periphery.
aren’t peripheries at every level, as in the LSC, as there are very few
lexical modifiers for PPs. The best examples are adverbs like right or
immediately, as in right behind the chair or immediately before the accident.
They would appear to be coreP-level modifiers. Predicative PPs share the
property of recursion with clauses in a limited way, as in The mouse ran out
from under the bed.
In some cases each member of the alternation occurs with a different verb,
as in The gang stole $5000 from the bank vs. The gang robbed the bank of $5000.
Non-predicative adpositions cannot be stacked the way predicative adposi-
tions can be, as in the earlier example out from under the bed.
The distinction between predicative and non-predicative adpositions is
very important for the analysis of many grammatical phenomena cross-
linguistically.
(12′) a. The notorious [God is dead] continental philosophers are mostly dead.
b. My grandson likes to give me the triumphant [who’s the boss now, silly old
grandpa] sly wink frequently.
Frau ‘the woman very proud of her daughter’ is presented in Figure 1.14. Two
points need to be mentioned. First, degree modifiers like sehr ‘very’ do not
head phrases and therefore are not in an MP. Second, the structure of RPs,
including the status of definite articles like die ‘the’ will be discussed in the
next section.
Like the predicative PP, the MP has one less layer than the LSC, and there
are no operators modifying it. There are peripheries at the nuclear level for
degree modifiers (illustrated) (see also Fleischhauer 2016) and at the core
level for manner adverbial modifiers, for instance the rapidly contained wild-
fire, the beautifully embroidered dress, the carefully constructed maze.
In the English translation of this RP, the MP, which is a reduced relative
clause, follows the head noun; when the MP consists of a participle, some,
but not all, may occur either pre-head or post-head, for example the murdered
man vs. the man murdered, the running man vs. the man running, the embroidered
sweater vs. *the sweater embroidered (but the sweater embroidered by Mary).
It was pointed out that the unusual phrasal modifiers in (12) cannot be
considered some kind of exotic relative clause or the like, as the examples in
(12′) show. What is so unusual about them is that they exhibit asymmetrical
embedding, that is, a larger unit (clause, sentence) is embedded in a smaller
unit (nucleus); this will be discussed in Section 1.3. The structure of the first
RP in (12a) is given in Figure 1.15; the structure of the RP with the sentence
as modifier in (12b) can be found in Van Valin (2008: 173).
The fact that the embedded clause is under the nucleus rather than being
under the MP node follows from the examples in (12′), in which the phrasal
modifier occurs in the middle of a series of modifiers. The structure of RPs
with multiple modifiers, for example the very expensive beautifully embroidered
context and the physical environment. These distinctions derive from the
proposals in Rijkhoff (1990, 2002). The primary nuclear operator in the
clause is aspect, which is realized prototypically by the imperfective–
perfective opposition, and a number of linguists, for example Jackendoff
(1990) and Talmy (2000), have argued that the imperfective–perfective con-
trast for verbs has, as its analogue among nouns, the count–mass distinc-
tion, with mass and imperfective, on the one hand, and count and
perfective, on the other, going together. If one needs to render a mass noun
(e.g. water) countable, languages make use of measure phrases or numeral
classifiers, for example two glasses/bottles/drops of water. A related use of
nominal classifiers is to specify the relevant semantic properties of a refer-
ent. Becker (1975), for example, gives the following Burmese examples of
classifiers contributing to the interpretation of an RP.
(51) a. myiɂ t yaɂ
e ‘river one place’ (e.g. destination for a picnic)
river one CL
b. myiɂ t tan
e ‘river one line’ (e.g. on a map)
c. myiɂ t hmwa
e ‘river one section’ (e.g. a fishing area)
d. myiɂ t ′sin
e ‘river one distant arc’ (e.g. a path to the sea)
e. myiɂ t θwɛ
e ‘river one connection’ (e.g. tying two villages)
f. myiɂ t ′pa
e ‘river one sacred object’ (e.g. in mythology)
g, myiɂ t khu′
e ‘river one conceptual unit’ (e.g. in a discussion of rivers
in general)
h. myiɂ t myiɂ
e ‘river one river’ (the unmarked case)
This nuclearR operator will be termed nominal aspect, and it covers the
count–mass distinction and the use of classifiers, numeral or nominal.
The core-level operators in the RP are concerned with quantity and quan-
tification. The primary operator notion at this level is number. It is unques-
tionably closed class and grammatical. Less clear-cut is whether numerals
and quantifiers are operators belonging to this domain. In previous work in
RRG, numerals were assumed to be coreR operators, but there are significant
problems with that analysis. First, numerals are not a closed class; in fact,
they are an infinite open class. Second, grammatically they behave like
lexical items, not grammatical morphemes. For example, they can function
as arguments in elliptical contexts and sometimes as part of the predicator
in the nucleus; number cannot.
are relevant here are definiteness and deixis. Definiteness covers several
related oppositions: referential vs. non-referential, specific vs. non-specific,
and definite vs. indefinite. These distinctions are canonically expressed by
articles, but it is well known that many languages lack articles. Neverthe-
less, these notions are relevant in article-less languages and are expressed by
word order, demonstratives, and/or the numeral meaning ‘one’ (Balogh et al.
2020), for example. Lakhota has a particularly rich system of articles, with
two definite articles, and an astonishing number of indefinite articles (see
Ullrich 2011: 810–815). The central distinctions are illustrated in the
following dialogue (Ullrich and Black Bear 2018: 108).
in second position, that is, Bata ba ang tumawa?’Is the one who laughed a
child?’ (cf. (7)).
In Figures 1.14–1.16 and 1.18 the modifiers expressing properties or char-
acteristics of the head of the RP occur in the nuclear periphery, which is the
locus of restrictive modification in the RP. Non-restrictive modification, on
the other hand, is associated with the RP-level periphery. In simple sen-
tences this is primarily realized through appositive modifiers, for example
Fred Jones, a lawyer, was arrested by the FBI on terrorism charges. In complex
sentences involving relative clauses, restrictive relative clauses occur in the
nuclear periphery and non-restrictive ones in the RP-level periphery. This
(54) a. I called the woman that I met on the train yesterday, who turned out to
be a potential investor, to set up a meeting with our company.
b. *I called the woman, who turned out to be a potential investor, that I met
on the train yesterday to set up a meeting with our company.
verb with a second oblique argument such as sit on, arrive at, pray for; a
three-argument template, which can accommodate three RPs (Mary gave
Chris a present), two RPs and a PP (Mary gave a present to Chris) or one RP and
two PPs (Sam ran from the school to the fire station). These possibilities highlight
the advantage of having underspecified templates; if the RPs or PPs were
specified in the core templates, then many more would be needed. Because
English word order is so rigid, especially with respect to the position of the
nucleus, it is possible to include the nucleus in the core template. That
would not be possible in German or Croatian, for example, because the
nucleus can occur in a variety of positions within the clause. The other two
core templates are for non-verbal predicators, which require an auxiliary,
normally be, as discussed in Section 1.2.3.3. The one-place template is for
sentences like The beer is cold or The weather is bad, and the two-place template
is for sentences such as Mary is happy about the election results or Sam is upset
with Sally. Below them are two operator projection templates, the first with
the obligatory operators in English, tense and IF, and the second one adding
an aspectual operator, which could be either progressive or perfect. In the
upper right are the templates for PPs, both predicative and non-predicative,
and there are three RP templates: the first two are for pronouns and proper
nouns, which do not take operators or modifiers, and accordingly lack a
layered structure. The third template is for common count nouns; the
obligatory operators, definiteness and number, are included. Finally, below
them are the most frequent ‘extra-spinal’ positions, the PrDP and the PrCS,
as well as the adjunct and modifier hosting peripheries and the MP. If the
sentence does not have any extra positions (PrDP, PrCS), then the ‘cap’ at
the bottom is employed.
The templates are stored in the SI, and when a semantic representation is
to be mapped to a possible sentence structure, the syntactic representation
is created by assembling the pieces from the SI according to the syntactic
template selection principles in the linking algorithm (see Section 1.6.1). The
construction of the syntactic structure for the sentence in Figure 1.6 is
presented in Figure 1.21.
The arrows indicate how the templates fit together; they do not consti-
tute a ‘derivation’ of the sentence. There is a PrDP, and it contains a PP
with a proper noun. The CLAUSE node in the PrDP template joins with the
CLAUSE node in the PrCS template, and the PrCS hosts a wh-word, what,
which is a type of pronoun. The core node in the PrCS template merges
with the core node of one of the core templates. The semantic representa-
tion of the predicator is the basis for selecting the appropriate core tem-
plate, as will be seen clearly when the linking algorithm is introduced later
in the chapter, and so one would expect a core with two core argument
slots, since eat is a transitive verb. However, one of the arguments of eat is a
wh-expression, and the default situation is for it to appear in the PrCS;
consequently a one-place core is selected. The appropriate operator
Figure 1.22 Layered structure of the word following Everett (2002) and English refusals
and the plural marker is an element in the corew. The difference between
the plural -s in English cats and the possessive clitic -’s in cat’s is reflected in
the LSW: cats would be [WORD [COREw [NUCw /kæt/] {PL}]], while cat’s would be
[WORD [COREw [NUCw /kæt/]][{POSS}]]. As with the LSPP and LSMP, there are no
operators modifying the layers of the word, hence there is no operator
projection in Everett’s model of the LSW. (See Cortés-Rodríguez (Chapter 8)
for a different view.)
1.2.7.2 Head-Marking
The visibility of inflectional morphology to the syntax is perhaps most
clearly observable in head-marking languages. RRG has been concerned with
head-marking grammar since its inception, since Lakhota, one of the
‘founding languages’ of the theory (Section 1.1), is consistently head-
marking. Van Valin (1977) presented an initial attempt at an analysis of
head-marked clause structure, followed by a more detailed account in Van
Valin (1985).27 The basic phenomena are illustrated in (55).
The verb kat’Á ‘beat to death’ (typically with an axe, club, baseball bat, etc.)
is made up of the stative verb t’Á ‘die, be dead’ and the instrumental prefix
ka- ‘by striking’; a more literal translation would be ‘cause to die by
striking’.28 In (55a) there is an object (undergoer) RP zuzéča kiŋ ‘the snake
(s)’, which is cross-referenced on the verb by the third-person plural animate
undergoer prefix wičhá-. The subject (actor) is signalled by the first-person
singular actor prefix wa-; no independent pronoun is required. There is no
case marking on the RP; it has the same form in (55a) as undergoer as it has
in (55b) as actor. The difference in interpretation derives from the different
ways it is cross-referenced on the predicator. In (55a′) the verb constitutes
the entire sentence, and it is with respect to the analysis of examples like
this that theories differ in important ways. Many posit null pronouns
occupying the same positions in the syntax as full RPs which trigger the
agreement/cross-reference on the verb. They are in effect treating these
languages as being dependent-marking with a lot of agreement morphology
(see Van Valin (2013) for detailed discussion). Such an analysis is impossible
in RRG because of the concreteness condition on syntactic representations
introduced in Section 1.1. Rather, as it has since Van Valin (1977), RRG
recognizes the bound affixes on the predicator as being the core arguments
that the syntax is sensitive to. An example of the syntax ‘seeing into the
inflectional morphology of a word’ is the obligatory control construction in
(55c). The first-person singular actor prefix on the linked verb, in this case
wa-, must be omitted, and the actor of iyútȟA ‘try’ must be interpreted as the
actor of the linked verb as well. If both verbs have actor marking, the result,
as in (55c′), is ungrammatical. No RPs or independent pronouns are
required. Only the bound argument markers on the predicator matter for
this construction. Moreover, the construction is oblivious to whether the
Figure 1.23 The structure of wičháwakat’iŋkte ‘I will beat them to death’ in (55a, a′)
The accusative clitic lo (singular) ~ los (plural) cannot co-occur in the same
clause with an overt object RP; they are in complementary distribution, as in
ˇ
Chichewa and Kikuyu. If both are to appear in a single utterance, the RP
must be in a detached position, either before or after the clause. The subject
marker on the finite verb or auxiliary, on the other hand, can co-occur with
an RP in the core or refer like a pronoun if there is no subject RP. Rather
than analysing these morphemes as sometime being agreement and some-
time being pronominal, Van Valin (2013) argued that they should be con-
sidered to be pronominal anaphors: they can be bound locally by an
argument, like an anaphor, or they can refer independently, like a pronoun,
ˇ
when not bound locally by an argument. In Chichewa, Kikuyu and Spanish,
the subject affixes are pronominal anaphors, while the object markers
are pronouns.
An interesting twist is provided by the dative clitics in Spanish: they are
pronominal anaphors, like the subject markers, and unlike the accusative
clitics, as (58) shows.
In (58a) the dative clitic co-occurs in the core with a dative RP, indicating
that it can be bound locally. It can also refer independently, as in (58b), and
(58c) shows that the clitic is obligatory with the verb gustar ‘like’. The wh-
question in (58d) shows that it can also be bound within a simple clause,
which is predicted by the pronominal anaphor analysis.
ˇ
Lakhota is different from Chichewa, Kikuyu and Spanish in that all of the
argument markers, not just the subject markers, are pronominal ana-
phors. They are bound within the clause when there are RPs, and when ˇ
there are none, they refer like pronouns. The contrast between Chichewa
and Kikuyu, on the one hand, and Lakhota, on the other, is significant,
because all three are consistently and thoroughly head-marking, yet they
differ in terms of the binding properties of their argument markers. This is
clearly a parameter of cross-linguistic variation in this type of language.
Another parameter of variation has been proposed in Bohnemeyer et al.
(2016) based on data from Yucatec Maya, another thoroughly head-marking
language. They argue that in Yucatec, RPs are in the core, not in ECSs as in
Lakhota. While they present their conclusions as tentative rather than
definitive, they are nevertheless very intriguing. There are at least two
ways to think about this. On the one hand, it could be taken as evidence
that the bound argument markers do not necessarily saturate the valence
slots in the core completely, as in Figure 1.24, but rather the RP and the
affix together satisfy them. Alternatively, this could be viewed as being
analogous to the variation in the distance between the antecedent (binder)
of a reflexive anaphor and the anaphor itself. English and German permit
only a short distance (at most a core) between the two, for example Nancy
bought a new dress for herself but not *Nancy told Bill to buy a new dress for herself,
whereas this sentence would be perfectly grammatical in Icelandic. With
regard to the status of RPs in the clause, the pronominal anaphors in
ˇ
Spanish, Chichewa, Kikuyu and Yucatec permit the antecedent to be
within the core, whereas Lakhota does not. This constraint on the binding
of pronominal anaphors in Lakhota does not affect reflexive constructions,
because they do not involve any binding; they are purely lexical (Van Valin
and LaPolla 1997, §7.5.1).
Thus, two parameters of typological variation among consistently head-
marking languages have been identified, and this highlights the need for
serious investigation of the typology of head-marking languages, as well as
double-marking languages, which have not been discussed here. (See Van
Valin (2005: 18–19) for some preliminary remarks.) As Nichols commented in
her seminal 1986 paper (116):
The term ‘complex sentences’ covers two distinct albeit related phenom-
ena, namely, clause linkage, which involves the combining of
predicator-based units, on the one hand, and complex RPs, which
1.3.1 Juncture
The answer to the question ‘what are the units of clause linkage’ is
simple and straightforward in RRG: the units of the LSC. This is summarized
in (59).
In (61a) the clause contains two cores, Hans hat versucht and die Tür aufzuma-
chen, while in (b) the two cores are Maria hat Hans überredet and das Auto zu
waschen. Clausal junctures have multiple clauses in a single sentence, as in
(62a), and sentential junctures have multiple sentences in a single discourse
unit, which is labelled here as ‘Text’, as in (62b).
(62) a. I’m aware Sam needs money to buy his wife a new car, but why did he
blow all that money betting on horses?
b. As for Germany, the response to the Covid-19 pandemic has been
surprisingly problematic, and as for the EU as a whole, the situation is grim.
In (62a) there are two clauses in the sentence, each with its own IF
operator; the first clause is an assertion and the second a question. In (b),
each clause has its own frame-setting topic in the PrDP, and that means each
of the conjuncts is a sentence, not just a clause.
It is possible to have all four juncture types in a single utterance, as
in (63).
(63) Poor Sally, she tried to fix the broken picture frame and have it done when
Fred got home, but the glue she tried, it didn’t hold the pieces
together properly.
The first sentence in the sentential juncture contains a clausal juncture, and
the first clause in it contains a core juncture (tried to fix), while the second
clause houses a nuclear juncture (have it done), and a temporal adverbial
clause (when Fred got home) modifying the core containing the nuclear junc-
ture. The second sentence contains a nuclear juncture (hold the pieces
together), which, like the one in the first sentence, is an example of an
English resultative construction, like wipe the table clean, paint the fence white.
A rough approximation of the constituent projection of (63) is given in
Figure 1.25. Details of phrases are not spelled out, and the conjunctions
are not represented.
Figure 1.25 The constituent projection for the multi-juncture utterance in (63)
1.3.2 Nexus
The structural relations between units in clause linkage are termed nexus
relations in RRG. Traditionally, two nexus relations are recognized,
namely, coordination and subordination. With respect to clauses, coordin-
ation means that the two clauses are independent of each other and each
can stand on its own as an independent utterance. It is a relation between
two wholes. Subordination, on the other hand, is a part–whole relationship,
in which the subordinate clause functions either as an argument of the
matrix clause or a modifier of it in some way; subordinate clauses cannot
stand on their own as independent utterances.
RRG has introduced two innovations with respect to nexus relations. First,
it applies them to the four layers of the clause: nucleus, core, clause and
sentence. Hence there is a contrast between the two traditional nexus types
at each of the four levels. Second, it posits a third nexus type,
cosubordination, which is a kind of dependent coordination.30 The
dependence is operator sharing: the linked unit must share an operator at
the level of juncture with the licensing unit. For English clauses, the relevant
operators are tense and IF. It is crucial, first and foremost, to distinguish
operator sharing from operator identity. This is illustrated in (64) and (65).
(64) a. Bill bought a new car yesterday, and he sold his old one, too.
b. Bill bought a new car yesterday, and did he sell his old one, too?
c. Bill bought a new car yesterday, and he will sell his old one tomorrow.
d. Bill bought a new car yesterday, and will he sell his old one tomorrow?
e. Did Bill buy a new car yesterday and will he sell his old one tomorrow?
(65) a. Bill bought a new car yesterday and sold his old one, too.
b. Did Bill buy a new car yesterday and sell his old one, too?
b′. *Did Bill buy a new car yesterday and sold his old one, too?
c. Bill bought a new car yesterday and will sell his old one tomorrow./*?
d. *Did Bill buy a new car yesterday and will sell his old one tomorrow?
In (64a) both clauses are past tense, and both have declarative IF; this is a
case of operator identity, not operator sharing. The two clausal operators
happen to be the same, but they are not obligatorily the same. This is shown
in (64b), where the tense is the same in both but IF differs, in (64c) where
tense differs but IF is the same, and in (64d) where both are different in each
clause. Both clauses can be questions, as in (64e), but the interrogative IF
must be indicated separately in each clause. There are no dependencies,
operator or otherwise, between the two clauses. Coreference between Bill
and he is not obligatory, despite being strongly preferred; in a context in
which Sam is the main discourse topic and the issue of car selling and
buying, he could be interpreted as referring to Sam rather than Bill.
This contrasts sharply with the examples in (65), which differ from those
in (64) in that the ‘subject’ of the second clause has been omitted under
identity with the ‘subject’ of the first clause.31 The sentences in (64a) and
(65a) look superficially similar, since they both have the same tense and IF in
both clauses, but they are not the same. In (65) the IF must be the same in
both clauses: in (65b) the tense operator is at the beginning of the clause
signalling interrogative IF, and the verbs in both clauses are bare infinitives,
which means they obligatorily share tense and IF in this interrogative
utterance. Sharing tense is not obligatory with declarative IF, as (65c) shows,
but the second clause cannot be interpreted as a question (hence ‘*?’), unlike
(64d). The examples in (65b′), (c) and (d) are all ungrammatical, because the
two clauses differ in IF, and this contrasts with (64b) and (d). Thus, the
sentences in (64a) and (65a) are both clausal junctures, but they differ in
nexus type: (64a) is coordination, while (65a) is cosubordination, because the
linked clause must have the same IF as the licensing clause and is therefore
dependent on it.
There is another important contrast illustrated here. Both (64) and (65) are
instances of conjunction, which is a kind of grammatical construction. Coord-
ination and cosubordination, as well as subordination, are abstract linkage
relations which are instantiated by a range of formal construction types. The
Mandarin construction in (60b) is an example of cosubordination at the
nuclear level; the two verbs dǎ ‘beat’ and sı̌ ‘die’ are followed by the perfective
aspect marker le, which obligatorily has scope over both of them. This is an
instance of an obligatorily shared operator at the level of juncture (perfective
aspect is a nuclear operator), and consequently the nexus type is cosubordina-
tion. The formal construction type is serial verbs, not conjunction. There is no
one-to-one correspondence between juncture and nexus combinations and
formal construction types. A useful analogy is to traditional grammatical
relations: they are abstract relations which are instantiated formally in a
variety of ways, namely, word order, case marking, cross-reference (as in
head-marking languages), agreement (as in ‘pro-drop’ languages) and even
tone (e.g. Maasai), but ‘subject’ is not necessarily identified or correlated with
a particular case or agreement or a position in the clause. The same is true of
clause-linkage relations. Coordination and cosubordination are not always
realized via conjunction, just like subordination is not always associated with
embedding, as will be shown later in this discussion.
This brings up a fundamental difference between the RRG theory of
complex sentences and other approaches. The following equivalence is
widely assumed: dependent ¼ embedded ¼ subordinate. Terms like ‘depend-
ent clause’, ‘embedded clause’ and ‘subordinate clause’ are used as syno-
nyms. In the RRG system, on the other hand, ‘dependent’, ‘embedded’ and
‘subordinate’ are not synonyms. Neither are ‘conjunction’ and ‘coordin-
ation’, despite common usage of them as synonyms. Coordination may be
realized in constructions which do not involve conjunction, as will be seen
in the discussion of core-level negation and at the end of this section.
Crucially, a dependent unit is not necessarily embedded or subordinate.
This can be seen clearly in (65); the second clause is dependent upon the
first clause for its IF value; if the first clause is an assertion, as in (65a, c),
then the second clause must be interpreted as an assertion as well, and if the
first clause is a question, as in (65b), then the second must be a question, too.
This is a particularly telling example, because the verb in the second clause
is a bare infinitive and accordingly the second clause is dependent on the
first clause for its IF value but also for its tense. If the tense in the first clause
changes, as in Will Bill buy a new car tomorrow and sell his old one, too?, the tense
interpretation of the second clause necessarily changes, too. Thus, the
second clause in (65) is dependent on the first clause, but it is neither
embedded nor subordinate.
The RRG treatment of sub-clausal junctures, specifically core junctures, is
also very different from that of other approaches. These include what are
known as ‘obligatory control’ and ‘raising’ or ‘exceptional case-marking’
constructions. These core junctures have an important property: they
require a shared core argument. This is illustrated in (66).
(67) a. Sam seems to have fixed the car. Shared argument: Sam ¼ core
argument of seem, actor of fix
a′. *It seems (for) Sam to have fixed the car.
b. Bill believes Tom to have fixed the car. Shared argument: Tom ¼ core
argument of believe, actor of
fix
b′. * Bill believes Tom (for) Sam to have fixed the car.
(69) a. Chris was stopped after running a red light. Peripheral ad-core
subordination
a′. *Chris was stopped after his/Mary’s running a red light.
b. Sally brushed her teeth before meeting Max for dinner.
b′. *Sally brushed her teeth before her/Sam’s meeting Max for dinner.
The passive test is not applicable to the structures in (66b) and (67), but the it-
cleft test is applicable; they all fail.
There is thus a clear contrast between try þ infinitive vs. try þ gerund with
respect to nexus. This illustrates an important point: a given predicator can
license more than one juncture–nexus combination (see Ohori, this volume,
Chapter 13, Van Valin 2005: 210).
The second difference also concerns the nexus type of these construc-
tions. The difference between the two non-subordinate nexus types is: if
there is a core operator (e.g. deontic modality) in the clause, it must have
scope over all of the cores in cosubordination but not in coordination.
This only applies when the primary participant (‘subject’) is the shared
argument between the cores, as in (74a–c); it does not apply to construc-
tions where the primary argument is not the shared argument, as in
(74d–e).
(75) a. Bill didn’t try to buy any/some flowers. (any > some)
b. Bill didn’t try to spend a red cent.
c. Bill didn’t persuade Sam to buy any/some flowers. (some > any)
d. ?Bill didn’t persuade Sam to spend a red cent.
(‘x > y’ means ‘x is preferred over y’.) In the cosubordinate examples, (75a, b),
negation in the first core can clearly license an NPI in the second. On the
other hand, in (75c, d) the licensing of the NPI in the second core is less
acceptable than in the first two examples. Core-level negation is more easily
shared across the cores in the cosubordinate rather than the coordinate core
juncture. This is a case of coordination not being realized by conjunction.
Figure 1.27 The structure of (65b): English clausal cosubordination with two shared operators
seen clearly in Figure 1.27, in which the superordinate clause nodes indicate
that tense and IF have scope over both clauses. There is no such superordin-
ate node in Figure 1.26, because each clause has its own tense and IF
operators. The structure of (65c) is interesting in that each clause has its
own tense but shared IF indicated by the position of the tense operator in
the first clause; it is given in Figure 1.28.
At sub-clausal levels the same patterns hold. The structure of (60b), a case
of nuclear cosubordination in Mandarin, is given in Figure 1.29.
In sub-clausal junctures, all higher layer operators are necessarily shared;
in a nuclear juncture like this one, all core and clausal operators have scope
over the nuclei, and similarly, in a core juncture all clausal operators have
scope over the cores.
Figure 1.28 The structure of (65c): English clausal cosubordination with one shared operator
(76) a. [CORE Larry believes [CLAUSE that Sam doesn’t like ice cream]]
b. [CORE [CLAUSE That Sam doesn’t like ice cream] surprised his friends]
(78) a. [CLAUSE [CORE Larry believed] [PERIPHERY yesterday] [CLAUSE that Sam doesn’t
like ice cream CORE] CLAUSE]
b. [CLAUSE [CORE It surprised his friends] [CLAUSE that Sam doesn’t like ice cream]
CLAUSE]
(79) a. Susan was upset with Paul for not washing the dishes after he got home
from the gym even though he had promised to do so.
b. Susan was upset with Paul for not washing the dishes even though he had
promised to do so after he got home from the gym.
There are two ad-subordinate clauses in these sentences: after he got home
from the gym, a temporal ad-core subordinate clause, and even though he had
promised to do so, a concessive ad-clausal subordinate clause. The two sen-
tences don’t have the same meaning: in (79a) the ad-core clause modifies the
main clause, so that the dishwashing was supposed to happen after the gym
due to a prior promise, whereas in (79b) the promise was made after the gym
and the time of the dishwashing is unspecified. The clause-level concessive
adjunct blocks the core-level temporal adjunct from modifying the main
clause, and it must be interpreted as modifying the core of the ad-clausal
clause, hence the difference in meaning.
It appears there are strong restrictions on what can modify the nucleus;
no nuclear-level PPs or ad-subordinate clauses modifying the nucleus have
thus far been encountered. Adjuncts in the nuclear-level periphery appear to
be restricted to adverbials.
There are some quite extreme instances of asymmetrical embedding; two
examples were given in (12), (12′) and Figure 1.15, in which whole clauses
and sentences are used as attributive modifiers in the MP, filling a slot that
would normally be filled by an adjectival modifier, possibly with a degree
modifier. A radically asymmetrical linkage can be seen in what Everett and
Kern (1997) and D. Everett (2008) term ‘intentional state constructions’ in
Wari’, the last viable Chapakuran language spoken in the Amazon in Brazil
and Bolivia. A simple Wari’ sentence is given in (80a) and an example of an
intentional state construction is given in (80b), taken from D. Everett (2008).
Basic word order is V-CL OS, and stress is on the last syllable of each word
(indicated by italics).
In (80a) the sentence begins with a verb followed by a clitic cluster cross-
referencing the direct core arguments, which is followed by an oblique core
argument followed by the two direct core arguments. In the intentional
state construction in (80b) the sentence begins with the wh-question ‘who
went to Guajará?’ followed by the clitic cluster and the two direct core
arguments. What is striking about (80b) is that the main clause has no verb
or predicate in it, and the slot for the nucleus is filled by the clause
representing the content of the speech. Everett presents a number of argu-
ments in favour of this analysis and shows that alternative analyses all fail
to capture the Wari’ facts. A simplified representation of (80b) is given in
Figure 1.30.
Here a clause is embedded in a nucleus, an extreme violation of the
tendency toward symmetrical embedding, but it is allowed by the theory
g. Intentionality
19. Propositional attitude: the expression of a participant’s attitude,
judgement or opinion regarding a state of affairs, e.g. Many people
believe that politicians are dishonest, Max considers Don to be a loser, Sports
fans want very much for their team to win.
20. Cognition: an expression of knowledge or mental activity, e.g. Bill
knows that he is in the wrong this time, Sally is thinking about how to fix the
problem with the microwave.
21. Emotion: an expression of the content of an emotional state, e.g. Sam
is happy/loves it that the pandemic is finally over. Larry is afraid/fears that the
corona virus will return.
h. Speech
22. Indirect discourse: an expression of reported speech, e.g. Frank said
that his friends are corrupt.
23. Direct discourse: The direct quotation of a speech event, e.g. Frank
said, ‘My friends are corrupt.’
i. Locational
24. Space: the spatial location of a state of affairs, e.g. Matilda lived her
whole life near where she was born.
25. Time: the temporal parameters of a state of affairs, e.g. Sally met Bob for
a drink after she left the office.
j. Circumstances
26. Reason: the motivation or cause for a state of affairs, e.g. The baby cried,
because it was hungry.
27. Conditional: an expression of what consequence would hold, given a
particular state of affairs, e.g. If it rains, we’ll have to stay home.
28. Concessive: the state of affairs specified in the main clause holds
unexpectedly, given the state of affairs specified in the subordinate
clause, e.g. Tim made it to work, even though it was snowing heavily.
k. Temporality
29. Simultaneous: one state of affairs is temporally coterminous with
another, e.g. Lisa yawned loudly while Dan was talking.
30. Sequential: one state of affairs follows another temporally, with or
without any temporal overlap, e.g. Marge finished breakfast, and then the
repairman was at the door.
31. Unordered: the temporal relation between states of affairs is
unspecified, e.g. Sam watched a football match, and Sally chatted with
her mother.
marked by to. Second, make and help require to in the passive voice, while
have and let do not occur in the passive voice with or without it.34 This is
illustrated in (82).
It has long been observed that for many English speakers the two versions
in (82a′) can have different interpretations (e.g. Quirk et al. 1972; Bolinger
1975; Dixon 1984, 1991): in the to-less version Sally participated in the
cleaning with Tom, whereas in the version with to she need not have
done any of the cleaning but rather could have just given Tom some
cleaning supplies. In other words, with the bare infinitive Sally’s helping
and Tom’s cleaning temporally overlap, implying that she did some of
the cleaning, while in the to-infinitive version they need not overlap. This
illustrates a claim made in Van Valin and LaPolla (1997, §8.4.2) that units
expressing temporally overlapping states of affairs are often not joined by
any kind of linkage marker. In (82a) the unspecified causing/letting action
is simultaneous with the caused/permitted action, that is, they overlap
temporally, and this is consistent with the bare infinitive complement
predicate. This is the marked case; the default is for infinitives in English
to be marked by to.
The juncture–nexus type of these examples is not readily apparent. Two
factors suggest that this construction might be a nuclear juncture. First, it
implicates unmediated causation with make, and in many languages con-
structions with an explicit causative verb are nuclear junctures, for example
French faire ‘make, do’ þ infinitive, Italian fare ‘make, do’ þ infinitive,
Jakaltek a’a’ ‘give’ þ infinitive. Second, nuclear junctures almost never have
any kind of linkage marker. In English, for example, there is a clear differ-
ence between Nancy seems happy and Nancy seems to be happy, such that the
former is a nuclear juncture, a complex predicate formed in the syntax, and
the latter a core juncture, which requires to (and be) and is not a
complex predicate.
Evidence against (82a) being a nuclear juncture comes from reflexiviza-
tion.35 English reflexivization has often been referred to as ‘clause bound’,
that is, the controller and the reflexive have to be in the same clause, but
that is not correct in terms of the LSC. In a simple sentence with a single
core, this is an accurate description. However, if the clause contains two
cores, the controller must be a semantic co-argument of the reflexive, as
illustrated in (83).
In (83a) Sam and himself are both semantic arguments of the verb injure, and
despite the occurrence of Sam in a different core, the reflexive is obligatory.
The same is true in (83b): Sam is the shared argument between manage and
injure, and therefore Sam is a co-semantic argument of injure with the
reflexive anaphor. In (83c) Donna is in the core headed by tell but is the
shared argument with injure, and therefore Donna and herself are both
semantic arguments of injure. In the final example, however, Sam and himself
are not semantic arguments of the same verb and are in different cores, and
consequently Sam cannot be interpreted as the binder of himself and the
plain pronoun him must be used to refer to Sam.
The crucial sentence is given in (84a), with possible structures as both
nuclear (84b) and core (84c) junctures.36
If the constructions in (82) are nuclear junctures, then the structure would
be as in (84b), and because Bill and himself are within the same core,
reflexive binding between Bill and himself should be possible. On the other
hand, if they are core junctures, then the structure would be as in (84c),
and because Bill and himself are in different cores and are the semantic
arguments of different verbs, reflexive binding should not be possible. The
sentence in (84) is quite ungrammatical with the reflexive but compatible
with a plain pronoun, which shows that it is a core juncture, not a
nuclear juncture.
The next question concerns the nexus type. Daughter subordination can
be ruled out for the reasons given in Section 1.3.2, for example *It was clean
the kitchen that Sally made/had/let/helped Tom, and it is clearly not an adjunct in
the ad-core periphery. It must, therefore be coordination or cosubordina-
tion. For the reasons discussed with reference to (74) in Section 1.3.2, the
deontic modal scope test does not apply to this construction, but the scope
of negation test in (75) is applicable. It is illustrated in (85).
(85) a. The teacher didn’t make the students buy anything/??something.
a′. The teacher made the students buy *anything/something.
b. No teacher had/let the students buy anything/??something.
b′. The teacher had/let the students buy anything/something.
c. The teacher didn’t help the students (to) buy anything/??something.
c′. The teacher helped the students (to) buy *anything/something.
Additional support comes from the MEP and the LSC (Bohnemeyer and
Van Valin 2017). Bohnemeyer and Van Valin argue that the RRG notion of
the (verbal) core is the natural unit for expressing macro-events. Crucial for
this discussion is the fact that macro-events can have at most one temporal
positional modifier (TPM) (excluding relative clauses, which are RP-internal
and may be a separate macro-event), as illustrated in (86).
The simple sentence in (86a) has the MEP, because it allows only one TPM,
not two as in (a′). In the other two examples, a case of core-level daughter
subordination in (b) and core coordination in (c), neither construction has
the MEP, because each of them has two TMPs. The MEP, they argue, is
preserved in complex sentences only in core cosubordination. In (74) and
(75) try þ infinitive constructions were shown to be cases of core
cosubordination, and in (87), it can be seen that they have the MEP.
(87) *Sam tried [TPM in the morning] to fix the bicycle [TPM in the afternoon].
The following examples show that the causative [2] constructions have
the MEP.
(88) a. *Sam made Bill [TPM in the morning] fix the bicycle [TPM in the afternoon].
b. *Sam had Bill [TPM in the morning] fix the bicycle [TPM in the afternoon].
c. *Sam let Bill [TPM in the morning] fix the bicycle [TPM in the afternoon].
d. *Sam helped Bill [TPM in the morning] fix the bicycle [TPM in the afternoon].
d′. ?Sam helped Bill [TPM in the morning] to fix the bicycle [TPM in the afternoon].
The first four examples do not allow more than one TPM, supporting the
conclusion that the nexus type is cosubordination. The last example, while
odd, is not as bad as the first four, and this is consonant with the claim that
the help þ bare infinite form involves temporally overlapping actions while
the help þ to-infinitive does not necessarily involve temporally overlapping
actions, which renders the two TPMs less unacceptable. There is a clear
contrast in acceptability between the examples in (86b, c) and those in
(88), and taken together with the contrast between (75) and (85), they
strongly support the analysis of these constructions as being instances of
core cosubordination.
It should be noted that they are very unusual examples of core cosubordi-
nation, for two reasons. First, they are the only cases of core cosubordination
in which the ‘subject’ of the licensing verb is not the shared argument with
the linked verb, and second, they are the only ones which take bare infini-
tives rather than to-infinitives in the linked core, which is a feature of a
tighter morphosyntactic bond between the cores. This suggests that this is
cannot easily stand on its own, for example ??I brushed little cat, ??Little cat
meowed, unless it is interpreted as a name. Subordination at the RP-level
involves non-restrictive modification, and non-restrictive relative clauses
are housed in the RP-level periphery.
At the coreR-level the relevant operators are negation and number. In
coordination, each unit in the construction has independent operators at
the level of juncture; higher level operators are by definition shared. In the
big dogs and little cat, the first unit is plural and the second singular, showing
operator independence at the level of juncture. This also works for negation,
as in No men and three women survived the crash. This is trickier in
cosubordination, because the number morpheme is a bound affix and is
required. So in English it is not possible to mark number on one unit and
have it apply to another unit, that is, *the dogs and cat-__, where cat is
interpreted as plural. This is, however, possible in Tagalog (Kolmer 1998),
because the plural marker is not an affix but a free, albeit clitic-like,
element, as in the equivalent to the example just given, ang mga aso at pusa,
in which it has scope over both of the linked units; the RP cannot be
interpreted to mean ‘dogs and cat’, only ‘dogs and cats’. There is operator
dependence across the linked units, hence it is cosubordination. With
respect to negation, there appear to be forms in which negation is obligator-
ily shared across the units (e.g. neither dogs nor cats, *either dogs nor cats,
??neither dogs or cats). Both types of subordination are found at this level.
Nominals derived primarily from verbs of intentionality and speech in (81)
can take propositional complements, e.g. the belief/claim/rumour/assertion/fear
that . . ., which would be coreR daughter subordination. Quantificational
lexical modifiers, namely, quantifiers and numbers, occur in the coreR-level
periphery, as discussed in Section 1.2.5.3.
The nuclearR-level operator is nominal aspect, which includes nominal
and numerical classifiers and the mass/count distinction. In the table it is
represented by English measure words, which are a kind of numerical
classifier. NuclearR coordination is straightforward, and nuclearR
cosubordination-1 is analogous to it in that it involves sharing nominal
aspect (numerical classifiers) across the two nuclei instead of each nucleus
having its own classifier. NuclearR cosubordination-2, on the hand, is analo-
gous to the nuclear juncture in (60b) from Mandarin, in which the two
verbal nuclei form a complex predicate which takes a single aspect operator.
It consists of compounding two nominal nuclei to form compounds, for
example wood chopper, air conditioner, fire station, knife sharpener, computer
repairman, cat litter, dog poop, which have a single value in terms of the
mass/count distinction, which is a nuclearR-level operator, despite not
having an overt morphological manifestation. It is, rather, what Whorf
called a ‘covert category’, one that is signalled by its morphosyntactic
‘reactance’. In these compounds, the value is determined by the final word
in the group, which means there is operator dependence across the two
nuclei: wood is a mass noun, chopper is a count noun, and wood chopper is a
count noun; knife is a count noun, sharpener is a count noun, and knife
sharpener is a count noun; cat is a count noun, litter is a mass noun, and cat
litter is a mass noun. The other nuclearR-level juncture–nexus type is periph-
eral ad-subordination, which, as discussed earlier, is the locus of restrictive
modification.
There is a further class of complex RPs which clearly involve juncture–
nexus distinctions but have proven difficult to characterize: they are nom-
inalization of jussive and other verbs which take an infinitive complement,
for example the (commander’s) order to the troops to attack the fortress, the
promise to the workers from the boss to increase their wages/the boss’s promise to
the workers to increase their wages, the request from the teacher to the students to
work quietly/the teacher’s request to the students to work quietly. There is a shared
argument in these constructions, just as in the corresponding verbal
predications discussed in Section 1.3.2, and accordingly these would be
non-subordinate core junctures. By virtue of the deontic modal scope test,
it was determined that the verbal predictions are core coordination, but
there are no comparable tests applicable to these complex RPs to distin-
guish the nexus types. It seems reasonable, then, as a preliminary analysis,
to categorize them as coreR coordination, based on the strong parallels
with the corresponding verbal predications. More research is needed on
these constructions.
1.4.1 Introduction
In Figure 1.1 there are two representations, one syntactic and the other
semantic, and the nature of the syntactic representation was explicated in
Sections 1.2 and 1.3. The spotlight in this section is the semantic representa-
tion, which contains two major parts: the system of lexical representation,
on the one hand, and the theory of semantic roles, on the other. The basics
of each will be presented, as more extensive discussion can be found in
Mairal Usón and Faber (Chapter 3) and Kailuweit (Chapter 4). Then some
extensions of these ideas will be presented.
(89) a. State: predicate′ (x) or (x, y), e.g. dead′ (rat) ‘The rat is dead’, know′ (Bill,
French) ‘Bill knows French’
b. Achievement: BECOME predicate′ (x) or (x, y), e.g. BECOME dead′ (rat) ‘The
rat died’, BECOME know′ (Bill, French) ‘Bill learned French’
c. Activity: DO predicate′ (x) or (x, y), e.g. DO dance′ (Sally), ‘Sally danced’, DO
eat′ (Bill, pizza) ‘Bill ate pizza’.
d. Accomplishment: ϕ CAUSE ψ (where ϕ is normally an activity predicate
and ψ an achievement predicate, e.g. [DO predicate′ (x)] CAUSE [BECOME
predicate′ (y) or (y, z)], e.g. [DO do′ (cat)] CAUSE [BECOME dead′ (rat)] ‘The
cat killed the rat’,
[DO do′ (Sally)] CAUSE [BECOME know′ (Bill, French)] ‘Sally taught French
to Bill’37
There are several appealing features of this approach. First, the main lexical
semantic content of verbs is represented by state and activity predicates
only, with change-of-state (achievement) and caused-change-of-state (accom-
plishment) verbs being derived via the addition of elements like BECOME
and CAUSE. Second, the state $ achievement $ accomplishment pattern is
directly reflected in the derivational morphology of verbs in many lan-
guages. Third, there is a set of syntactic and semantic tests for identifying
the class to which a particular use of a verb can be assigned (see Van Valin
2005, §2.1.1). One of the serious complications in the analysis of verb seman-
tics is that a given predicate, for example cool, can be used in more than one
way, as in The soup is cool (state), The soup cooled (achievement) and The ice cooled
the soup (accomplishment), and consequently what one is identifying in most
cases is the interpretation of a predicate in a particular sentence. The
question of how to analyse such a predicate is left open: one could claim
that there are three entries in the lexicon for the verb form cool (cool1 ¼ state,
cool2 ¼ achievement, cool3 ¼ accomplishment), or one could claim that there
is only one entry, cool ¼ cool′ (x), with the other two forms derived by means
of lexical derivational rules. RRG has always favoured the latter approach
(e.g. Van Valin 2012a).
The decomposition system was basically the same from 1984 through
1997, when Van Valin and LaPolla introduced a major revision of it. The
primary features of the revision involved returning to the original distinc-
tions as proposed in Vendler (1967)38 and factoring out causation as an
independent parameter: all of the Aktionsart types have causative and non-
causative versions. With respect to state predicates, more distinctions
among subtypes were recognized, following Schwartz (1993). Activity
predicates were marked with do′, in order to distinguish them from states
(e.g. see′ (Mary, child) ‘Mary saw the child’ [state] vs. do′ (Mary, [see′ (Mary,
child)]) ‘Mary watched the child’ [activity]). Change-of-state predicates went
from one category, achievements, to three: achievements (punctual
process accomplishment melt given above is PROC melt′ & INGR melted′,
and the question arises, if melted′ is a state, what is melt′? The system has
only two types of basic predicates, states and activities, and melt′ fits into
neither category. The same issue appears in the active accomplishment LSs
in (90a, b) with PROC consume′/create′ (y). Osswald notes that the intended
meaning in (90) is being.consumed′ and being.created′.
Processes are non-punctual changes of state or location, and therefore the
argument-bearing lexical predicate at the heart of the representation must
be either one of state or condition, or one of location. Many processes do not
have an incremental theme argument which is being consumed or created
or an incremental path; rather, they involve the position of the referent on
a scale. This is particularly true of what Dowty (1979) termed ‘degree
achievements’ such as cool, warm, widen, narrow, grow, melt, freeze, redden
(all [M]-intransitive), all of which assume a scale of some kind. For example,
there is a temperature scale with ‘hot’ at the top and ‘cold’ at the bottom, as
can be seen in the contrast between cool down/*up vs. warm up/*down. Osswald
notes that as a process the soup cooled means ‘the soup became cooler’, not
‘the soup became cool’. Hence the incremental theme of these verbs is
associated with movement on a scale rather than consumption or creation.
This can be represented as PROC becoming.higher/lower.on.[α]scale′ (x).
These representations carry over to their [M]-transitive causative
counterparts.
Nothing has been said about motion active accomplishments, which
involve an incremental path rather than an incremental theme. Crucially,
an incremental theme undergoes a change of state, but an incremental
path does not; it measures out the distance the x-argument covers, which
means the proposed predicate, covering.path.distance′, cannot be a predi-
cate of state or condition. It is, rather, a change-of-location predicate,
which can co-occur with motion activity predicates, as well as pure process
predicates like fall, as in He fell fifty metres into the water. In motion active
accomplishments there is an activity simultaneous with a process, the
former characterizing the motion itself and the latter expressing the
distance along the incremental path, whereas in the example with fall
there is an uncontrolled change of location in a specific direction covering
a specified distance before reaching a final location. Hence the LS for an
active accomplishment like stumble into the room or enter the room stumbling
should not be (90c) but rather [do′ (x, [move′ (x)]) ^ PROC covering.path.
distance′ (x, (y))] & INGR be-LOC ′ (z, x).41 Consequently, the LS for motion
active accomplishments parallels the LSs for creation and consumption
predications.42 On the other hand, the LS for the example with fall would
be [PROC moving.downward′ (x) ^ PROC covering.path.distance′ (x, (y))] &
INGR be-LOC ′ (z, x), where ‘moving′’ means ‘changing location’. Moving′
differs from do′ (x, [move′ (x)]), in that it necessarily does not involve any
effort on the part of the referent of its argument, is primarily associated
with direction rather than manner and is not subject to the agentivity
implicature, even with a human referent (see §1.4.3.1), whereas do′ (x, [move
′ (x)]) differs on all three of these characteristics.
The second issue raised by Osswald is what he calls the ‘and-then anomaly’.
He points out that representations like ‘PROC melt′ (x) & INGR melted′ (x)’
don’t accurately represent the situation. This one means ‘there was a process
of melting, and then after the process ended there was a transition to the
state of being melted’, but the state of being melted does not come about
after the process ended; rather, it is the final stage of the process itself.
Similarly, in active accomplishments like Sam devoured the pizza, it is not the
case that the activity of eating and the process of being consumed termin-
ated and then the pizza was consumed; the pizza was consumed at the
termination of the eating, not after it. A possible solution to this anomaly is
to introduce a function FIN ‘final stage of a process or action’ and change the
connector from & ‘and then’ to ^ ‘and simultaneously’, which is adapted
from Osswald’s proposal. Thus The ice cream melted would be PROC becoming.
lower.on.[solid$liquid]scale′ (ice cream) ^ FIN melted′ (ice cream), mean-
ing ‘the ice cream underwent a process of becoming lower on the “solid-to-
liquid” scale and the final stage is that the ice cream is melted’. For Sam
devoured the pizza, the LS would be do′ (Sam, [eat′ (Sam, pizza)]) ^ PROC being.
consumed′ (pizza)] ^ FIN consumed′ (pizza), meaning ‘Sam eats pizza and at
the same time the pizza undergoes a process of consumption whose final
stage is the pizza is consumed’.
The introduction of the FIN function offers a way to capture the contrast
between quantized and non-quantized changes of state. Following Beavers
(2013), Bentley (2019) argues that the previous decomposition system does
not express the difference between quantized change-of-state predicates
like die, which lexicalize a specific result state, and non-quantized change-
of-state predicates, which are of two types: those like melt and fill, which
entail only that a specific result state (e.g. being liquid, being full) might
but need not necessarily be reached, and those like widen or shorten which
have no specific final state. The following represents the result state
of the two subclasses of non-quantized change (D. Bentley, personal
communication):
This notation yields the necessary contrast with verbs of quantized change,
which entail the reaching of a specific final state (e.g. FIN dead′ (x)).
The final addition to the array of Aktionsart types is semelfactives
(Smith 1997), which are punctual events with no change of state or result
(91) a. STATES
The window is shattered. shattered′ (window)
John saw the picture. see′ (John, picture)
b. ACTIVITIES
The children cried. do′ (children, [cry′ (children)])
Carl ate pizza. do′ (Carl, [eat′ (Carl, pizza)])
c. ACHIEVEMENTS
The window shattered. INGR shattered′ (window)
The climber reached the summit. INGR be-at′ (summit, climber)
d. SEMELFACTIVES
Mary coughed. SEML do′ (Mary, [cough′ (Mary)])
Dana glimpsed the picture. SEML do′ (Dana, [see′ (Dana, picture)])
e. ACCOMPLISHMENTS
The ship sank. BECOME sunken′ (ship)
Mary learned French. BECOME know′ (Mary, French)
f. ACTIVE ACCOMPLISHMENTS
Carl ate the pizza. do′ (Carl, [eat′ (Carl, pizza)]) ^ PROC being.consumed′
(pizza) ^ FIN consumed′ (pizza)
Chris ran two miles to the park. do′ (Chris, [run′ (Chris)]) ^ PROC covering.path.distance′
(Chris, two miles) ^ FIN be-at′ (park, Chris)
g. CAUSATIVES
The dog scares the boy. [do′ (dog, Ø)] CAUSE [feel′ (boy, [afraid′])]
The submarine sank the ship. [do′ (submarine, Ø)] CAUSE [BECOME sunken′ (ship)]
The cat popped the balloon [do′ (cat, Ø)] CAUSE [INGR popped′ (balloon)]
The conductor flashed the light. [do′ (conductor, Ø)] CAUSE [SEML do′ (light, ([flash′ (light)])]
Felix rolled the ball. [do′ (Felix, Ø)] CAUSE [do′ (ball, [roll′ (ball)])]
Mary fed the pizza to the child. [do′ (Mary, Ø)] CAUSE [do′ (child, [eat′ (child, pizza)]) ^
PROC being.consumed′ (pizza) ^ FIN consumed′ (pizza)]
The causative LSs typically have an activity predicate as the first argument
of CAUSE,44 and it is quite common cross-linguistically for that activity to be
unspecified, as in all of the examples in (91). This unspecified activity
has heretofore been represented as ‘[do′ (x, Ø)]’. An alternative way of
representing this activity but with more semantic content is the
following: [do′ (x, [affect′ (x, y))] CAUSE [BECOME/INGR pred′ (y)]], which
signifies ‘x does something unspecified which affects y causing y to undergo
a change of state’. This has the advantage that it ties the unspecified causing
activity to the affected participant in the second LS in the causative LS. For
example, the LS for The burglar smashed the window would be ‘[do′ (burglar,
[affect′ (burglar, window))] CAUSE [INGR smashed′ (window)]’. Another alter-
native along the same lines is ‘[do′ (burglar, [act.on′ (burglar, window))]
CAUSE [INGR smashed′ (window)]]’. For the examples in (91g) any of the
three would be appropriate, but there are some interesting cases where
either of the latter two would be better. They involve predicative
prepositional phrases.
In Section 1.2.5.1 the distinction between predicative and non-predicative
PPs was introduced (see also Ibáñez Cerda, Chapter 10). Non-predicative
adpositions are oblique core arguments of the predicator in the nucleus of
the core, and they are assigned by rule or constructionally and are not directly
represented in the LS of the predicator. Predicative adpositions, on the other
hand, are predicates and are represented as such in the semantic representa-
tion of the sentence. In a sentence like Pam saw Felix in the library after the
lecture, the LS would be be-after′ (lecture, [be-in′ (library, [see′ (Pam, Felix)])]), in
which after the lecture (be-after′ (x, y)) and in the library (be-in′ (x, y)) are realized
as adjunct PPs in the core-level periphery. The first argument is the spatial or
temporal location of the event represented by the second argument. The
default situation is that non-predicative PPs occur in the core as oblique core
arguments and predicative PPs occur in a periphery as an adjunct.
There are, however, two situations in which this default distribution of
PPs does not hold. The first is the by-PP in a passive construction; it is a non-
predicative PP because it is constructionally motivated and is not repre-
sented in the LS, but it occurs in the core-level periphery. It is a case of a
causative morpheme like Lakhota ka- ‘cause by striking’. Second, the predi-
cate in an attributive LS can also occur in another type of LS. The adjective
black can occur as an attributive predicate, as in be′ (door, [black′]) The door is
black, an attributive modifier, as in like′ (Mary, [be′ (door, [black′])]) Mary liked
the black door, and also in a causative change of state LS, as in [do′ (fire, [act.
on′ (fire, door)])] CAUSE [BECOME black′ (door)] The fire blackened the door. In
this sentence black′ (door) is the result state of a causative accomplishment,
not an attribute, and accordingly the representation is different. Third,
attributive predicates can have second arguments, for instance Sam is proud
of his son, Nancy is angry at her husband, and the former would be represented
as be′ (Sam, [proud.of′ ([have.as.kin′ (3sgM, son)])]).
Adverbs have a simple LS, as they are one-place predicates, for example Bill
carefully closed the door slowly yesterday would be yesterday′ ([do′ (Bill, [careful′
(act.on′ (Bill, door))])] CAUSE [slow′ (BECOME closed′ (door))]). As this
example illustrates, adverbs may have the whole LS as their argument or a
subpart, depending on the level and scope of modification, as discussed in
Section 1.2.4.2.
The semantic representation of nominals does not treat them as predi-
cates, and their semantic representation is based on the qualia analysis
proposed in Generative Lexicon Theory (Pustejovsky 1995). The qualia them-
selves are represented in terms of the decomposition system used in RRG,
which differs from that used by Pustejovsky. See Van Valin (2005, §2.3) for
further discussion.
Operators have a place in the semantic representation of both RPs and
clauses. The RP operators and the semantic representation for Sally’s two red
parrots in Figure 1.18 are in (92).
(92) a. ⟨DEF ⟨DEIX PROX/DISTAL ⟨NEG ⟨NUM SG/DL/PL ⟨NASP COUNT/MASS ⟨[N]⟩⟩⟩⟩⟩⟩
b. ⟨DEF þ ⟨NUM PL ⟨NASP COUNT ⟨[have′ (Sally, [be′ ([be′ (parrot, [red′])]) [two′])])]⟩⟩⟩⟩
The operators over the clause are summarized in (93a) and the semantic
representation for The dogs may be barking at Sally’s two red parrots is given in
(93b). These semantic representations must ultimately be interpreted within
a formal semantic framework.
(93) a. ⟨IF DECL ⟨EVID Ø ⟨TNS PAST ⟨STA IRR ⟨NEG ⟨MOD ABLE ⟨DIR Ø ⟨ASP IMPF ⟨[LS]⟩⟩⟩⟩⟩⟩⟩⟩⟩
⟨DEF þ ⟨NUM PL ⟨NASP COUNT ⟨[dog (x)]⟩⟩⟩⟩
\
b. ⟨IF DECL ⟨TNS PRES ⟨STA PSBL ⟨NEG – ⟨ASP PROG ⟨[do′ (x, [bark.at′ (x, y)])]⟩⟩⟩⟩⟩⟩
/
⟨DEF þ ⟨NUM PL ⟨NASP COUNT ⟨[have′ (Sally, [be′ ([be′ (parrot (y), [red′])]) [two′])])]⟩⟩⟩⟩
relations of the IRH in (81) was originally suggested in Ohori (2001), and
Van Valin (2005: 207–208) used it to formalize the version of the IRH
assumed then. Before the revised system is presented, a few notes of
clarification are in order. First, ‘VERB ′ ’ stands for the class of predicates
of this type; thus, know′ is the verb know but KNOW ′ symbolizes the class of
cognition verbs. Causative [1] has the same LS as the lexical causative verbs
in (91g) and need not involve a causative verb like English make, German
lassen or French laisser, whereas causative [2] typically does feature an expli-
cit causative verb; causative [3] always involves two verbs, one of which may
be explicitly causative, such as English cause, force, or not, for example
English teach in teach Sam to fix bicycles. ‘pred′ (x, [LS . . . x . . .])’ means that
the participant denoted by x is involved in both the matrix and embedded
LSs, as in (66). ‘pred′ (x, [LS . . . y . . .])’ also signals that the participant denoted
by y is involved in both the matrix and embedded LSs but in a different way,
as in (67).
f. Perception
17. Direct PERCEIVE′ (x, [LS . . . y . . .])
18. Indirect PERCEIVE′ (x, [LS])
g. Intentionality
19. Prop. attitude BELIEVE′ ((x,) [LS])
20. Cognition KNOW′ (x, [LS])
21. Emotion FEEL′ (x, [LS])
h. Speech
22. Indirect discourse do′ (x, [say′ (x, [LS ⟨TNS . . . ⟩])])
23. Direct discourse do′ (x, [say′ (x, [LS ⟨IF . . . ⟩])])
i. Locational
24. Space be-LOC′ ([LS1], [LS2])
25. Time be-TEMP′ ([LS1], [LS2])
j. Circumstances
26. Reason [LS1] BECAUSE′ [LS2]
27. Conditional [LS1] [LS2]
28. Concessive [LS1] IN.SPITE.OF′ [LS2]
k. Temporality
29. Simultaneous [LS1] ^ [LS2]
30. Sequential [LS1] & [LS2]
31. Unordered [LS1] þ [LS2]
Because of this problem, it might be suggested that the LSs should reflect the
properties of the specific constructions. On this view, the LS for (95a) would
be something like be-at′ (1sg, [exist′ (knig-)]), which can be readily accommo-
dated in terms of the regular Russian linking and case assignment rules;
(96a) would be exist′ ([have′ (3sg, bayan)]), where the possessor can be linked
as an RP-internal possessor or linked via ‘possessor raising’ as an argument
in the core of the clause; and (97a) would include exist′ (libro), but it’s not
clear how the possessor is related to it, as it seems to be simply juxtaposed to
it (i.e. ([exist′ (libro)] þ babae)), which is unsatisfactory. Also unsatisfactory is
the semantic interpretation of these LSs, since they should mean the same as
their counterparts in the languages mentioned above.
In Latrouite and Van Valin (2014) a solution to this problem is pro-
posed: constructional meaning is represented by the same basic decom-
positional system as lexical meaning, but the notation for each is
distinctive. Lexical meaning continues to be represented as before, with
elements in the semantic metalanguage in boldfaceþprime (e.g. have′).
Constructional meaning, on the other hand, will be represented by small
capsþprime (e.g. have′). In building representations of constructional
meaning, the components INGR, PROC, BECOME and SEML have the same
function as in the representation of lexical meaning, but CAUSE must now
be treated differently in the two representational notations, because it can
be lexically realized, as in the examples in (91g), still symbolized by CAUSE,
or it can be constructionally induced, as in the examples discussed earlier,
symbolized by cause in small caps, for example [do′ (1sg, [beat′ (1sg, shé)])]
cause [INGR dead′ (shé)] for (60b) from Mandarin, [do′ (Mary, [paint′ (Mary,
door)])] cause [INGR black′ (door)] for Mary painted the door black. Accord-
ingly, the LS in (94a1) should have CAUSE*, which indicates both are
possible, and (94a2i) should have cause rather than CAUSE, since it typic-
ally does not contain a lexical coding of causation, unlike the causative [2]
relations in (94b6).
With respect to the possessive predications in (95)–(97), the LSs would be
be-at′ (1sg, [exist′ (knig-)]) ^ have′ (1sg, knig-) for Russian in (95a), exist′
([have′ (3sg, bayan)]) ^ have′ (3sg, bayan) for Guugu Yimidhirr in (96a), and
exist′ (libro) ^ have′ (babae, libro) for Tagalog in (97a). The Tagalog case is
the most interesting one, because, as noted earlier, there is no obvious way
to relate the possessor to the existential predication, and the constructional
LS supplies the connection. Latrouite and Van Valin (2014) give an analysis of
what are called ‘event existential constructions’ in Tagalog, which, they
argue, are a recent innovation and are related to possessive constructions,
roughly, ‘the woman has a book which was bought’ ) ‘there was an event of
book buying by the woman’. Their RRG account of the transition from
possessive to event existential construction crucially involves both the exist-
ential and the possession components in the LS. Thus, these representations
make explicit what is similar and what is different with respect to the
meaning of comparable constructions across languages.
An important feature of the RRG system is that the basic state and activity
predicates take no more than two arguments, and this entails that three-
place predicates must be composed of more than one basic predicate. The
primary way this is done is through forming causative LSs (see Table 1.4),
and the vast majority of three-place predicates are in fact causative.
The argument structure of a verb is the sum of the arguments of its
constituent state and activity predicates. For example, Bill knows algebra
would be know′ (Bill, algebra). Bill learned algebra would be BECOME know′
(Bill, algebra), with no change in the argument structure. Teach is a three-
place predicate, as in Mr Smith taught Bill algebra, and as predicted, it has a
causative LS, namely, [do′ (Mr Smith, Ø)] CAUSE [BECOME know′ (Bill,
algebra)]. The argument structure of teach is the combined argument struc-
tures of the constituent state and activity predicates, namely effector, cog-
nizer and content. This has important theoretical consequences, as
discussed in Kailuweit (Chapter 4) and Van Valin (2005, §2.4.1).
The actor is the most agent-like argument in the LS, namely the leftmost,
while the undergoer is the most patient-like argument in the LS, namely the
rightmost. There is an interesting asymmetry between actor and undergoer
here: the actor is always the leftmost argument in the LS, whereas the under-
goer as the rightmost argument is, for some verbs in some languages, only the
default situation. This follows from an asymmetry in lexicalization. The
meanings of the macroroles are very general: the actor is the participant
who is responsible for the state of affairs, in the sense that there can’t be an
event of running without a runner, there can’t be a cognitive event without a
cognizer, a perceptual event without a perceiver, etc., whereas the undergoer
is the most affected participant in the state of affairs. Languages lexicalize
responsibility but not necessarily affectedness. As examples of variable affect-
edness, consider the pairs of examples in (100).
(100) a. Marie taught French to the students (but they didn’t learn a word of it).
a′. Marie taught the students French (?but they didn’t learn a word of it).
b. The workmen loaded the hay on the truck. (all of the hay, the truck may or
may not be full)
b′. The workmen loaded the truck with the hay. (the truck is full (preferred),
all of the hay may or may not have been loaded)
c. The company shipped the package to Henry (but he never received it).
c′. The company shipped Henry the package (?but he never received it).
The default interpretation of (101a) is that shown in (b), not the one in (c),
but both are possible. If the actor of kill is an agent, understood as the wilful,
The picture is very different here. The actor of murder must be an agent as
described above, and consequently (b) is redundant, (c) is a contradiction,
and (d) is nonsensical. This is different from the actor of kill, which may,
but need not be, an agent. Rather, following Holisky (1987) and Van Valin
and Wilkins (1996), the actor of kill is an effector, which can be interpreted
as an agent if it is human and there is no evidence to the contrary.
Accordingly, the default interpretation of (101a) is (101b), as noted above,
in which the adverb intentionally specifies that the actor is agentive. In (c)
the adverb inadvertently constitutes evidence to the contrary and blocks
the agent implicature, which does not apply in (d) due to the non-
human effector.
Three points deserve mention. First, the agent implicature applies only to
effector arguments. The human actor of receive or see is not construed as an
agent. Receive has a non-volitional recipient-actor and contrasts with accept,
which has a volitional recipient-actor. See has a non-volitional perceiver-
actor, but watch and look at, which are activity verbs of directed perception
(see fn. 50), are subject to the agent implicature. Second, agent is fundamen-
tally different from all other thematic relations in that it is always an
overlay on top of another thematic relation, most commonly effector. Third,
languages differ in the extent to which they lexicalize agentivity. English
seems to rely on the agent implicature, as does Tsova Tush (Caucasian;
Holisky 1987), whereas in Japanese the human actor of a transitive verb
must be interpreted as an agent (Hasegawa 1996), and the agent implicature
plays no role.
There is also an implicature with undergoer. The first thing to point out is
that in English the undergoer with an active voice transitive verb is the
direct RP immediately following the nucleus, and this means that Sam is the
undergoer in (103a) and the book is the undergoer in (a′). This does not
correspond to the traditional analysis that claims that the book is the direct
object in both sentences and Sam is likewise the indirect object in both.
Evidence comes from passivization, which has the undergoer as the subject
in English. There are two candidates for the passive form of (a), as in (103b, b′);
there is only one possible passive of (a′), The book was given to Sam by Mary.
(104) a. Mary gave Sam the book (*but he never received it).
a′. Mary gave the book to Sam (*but he never received it).
b. #Mary gave Boston the book.
b′. #Mary gave the book to Boston.
c. Mary sent Sam the book (?but he never received it).
c′. Mary sent the book to Sam (but he never received it).
d. ?Mary sent Boston the book.
d′. Mary sent the book to Boston.
Valence Macroroles
snow 0 0 M-atransitive
faint 1 1 M-intransitive
limp 1 1 M-intransitive
drink [ACT] 1 or 2 1 M-intransitive
drink [ACTACC] 2 2 M-transitive
smash 2 2 M-transitive
show 3 2 M-transitive
give 3 2 M-transitive
There are two principles here, the first one governing the number of macro-
roles a verb takes based on the number of arguments specified in its LS, while
the second governs which macrorole an M-intransitive predicate takes. The
crucial idea in (105b) is that actor macroroles are tied to the presence of an
activity predicate in the LS of M-intransitive verbs. There is an important
asymmetry between these two principles: there are very few exceptions to
(105b) concerning the identity of the single macrorole, but there are numer-
ous exceptions to (105a), which reflects the fact that the same basic semantic
content can be expressed more than one way. An example from the domain of
possession in English and German is own and besitzen (M-transitive) vs. belong
(to) and gehören (M-intransitive). All of these verbs have basically the same LS,
have′ (x, y), and own and besitzen follow the default principle in (105a). On the
other hand, belong (to) and gehören do not, and this can be indicated by
specifying the macrorole number in the LS, namely, have′ (x, y) [MR 1].
Nothing else needs to be specified; all of the morphosyntactic properties of
the four verbs follow from the linking algorithm, given their LS and macro-
role number, as will be shown in Section 1.6.
(106) a. kill [do′ (x, [act.on′ (x, y)])] CAUSE [INGR dead′ (y)]
b. receive INGR have′ (x, y)
c. own have′ (x, y)
d. belong (to) have′ (x, y) [MR 1]
e. arrive INGR be-at′ (x, y)
f. go do′ (x, [move.away.from.ref.point′ (x)]) ^ PROC covering.path.
distance′ (x, (y)) ^ FIN be-LOC ′ (z, x)
g. seem seem′ (x, y) [MR 0]53
h. see see′ (x, y)
i. watch do′ (x, [see′ (x, y)])
j. show [do′ (w, Ø)] CAUSE [INGR see′ (x, y)]
k. run do′ (x, [run′ (x)])
l. drink do′ (x, [drink′ (x, y)])
m. melt BECOME melt′ (x)
n. afraid feel′ (x, [afraid′ (y)])
The lexicon is a vital component in the RRG system, and it is more than
just a storehouse for words and morphemes. It can be divided into two
components, which may be metaphorically labelled ‘the warehouse’ and
‘the workshop’. The workshop is the location where lexical and derivational
rules apply, and therefore it is where the LS for the semantic representation
is assembled, among other operations. Crucially, the lexical items to fill the
argument positions in the semantic representation are selected at this point.
More will be said about this in Section 1.6.
Figure 1.33 The constituent, operator and information structure projections of the clause
The heart of the RRG system is the linking algorithm which maps between the
semantic and syntactic representations. It is bidirectional: it maps the seman-
tic representation into the syntactic representation, and it maps from syntax
to semantics as well. This reflects the fact that language users are bidirec-
tional. A speaker goes from a message to be conveyed to the formal packaging
of it which is to be uttered (semantics to syntax), while a hearer analyses the
formally packaged message and gives it an interpretation (syntax to seman-
tics). As discussed in Chapter 19, RRG is not a neurocognitive processing
model but has the potential to form the basis for one.
Specific aspects of the linking algorithm are discussed in depth in differ-
ent chapters (e.g. grammatical relations in Chapter 5, case assignment in
Chapter 7, and argument structure alternations in Chapter 6). The purpose
of this section is to give an overview of the system as a whole. Each direction
of mapping will be presented separately, starting with semantics-to-syntax
linking. The linking algorithm is governed by a very important principle,
the Completeness Constraint.
Steps 1 and 2 take place in the lexicon, specifically in the ‘workshop’. The
central move in step 1 is the selection of the predicator in the nucleus and
its LS. The argument positions in it must be filled by referring expressions,
and the choice is determined by the activation status of the referents of the
referring expressions; see Chapter 11. In addition, any adjunct LSs must also
be incorporated into the representation, along with the relevant operators.
As an example, the situation to be described is one in which a woman draws
a man’s attention to something in a work situation subsequent to an online
meeting. The verbal predicator selected is show, and the referring expres-
sions to fill the argument positions are Linda, the salesman, and what. In
addition, there are two predicative prepositions, after and in, which take
one referring expression each, namely the meeting and the office, respectively;
the other argument position is filled by the LS of show. Finally, the operator
values are determined, specifically past tense and interrogative IF. The result
is the representation in (109).
(109) ⟨IF INT ⟨TNS PAST [be-after′ (meeting, [be-in′ (office, [[do′ (Linda, Ø)] CAUSE
[INGR see′ (salesman, what)]])])]⟩⟩
Step 2 is to assign macroroles, and while Linda must be the actor, the speaker
has a choice with respect to the assignment of undergoer. The default is that
what would be the undergoer, and that choice will be reflected in the
following discussion. The result of Step 2 is (110).
(110) ⟨IF INT ⟨TNS PAST [be-after′ (meeting, [be-in′ (office, [[do′ (A: Linda, Ø)] CAUSE
[INGR see′ (NMR: salesman, U: what)]])])]⟩⟩
arg of DO > 1st arg of do′ > 1st arg of pred′ (x, y) > 2nd arg of pred′ (x, y) >
arg of pred′(x)
In both languages, the default word order is for the dative RP to be first in
the core, and the finite verb agrees with the nominative RP. If we put these
clauses into a conjunction reduction construction, as in (113), the different
status of the dative RPs is clear (Rögnvaldsson 1982; Zaenen et al. 1985).55
b. Ergative constructions
1. Assign absolutive case to the lowest-ranking macrorole.
2. Assign ergative case to the other macrorole.
3. Assign dative case to non-macrorole arguments (default).
When there is only one macrorole in a core, it counts as the highest ranking
for (114a) or the lowest ranking for (114b). The English preposition assign-
ment rule that is relevant to the example in (109) is given in (115).
(116) ⟨IF INT ⟨TNS PAST [be-after′ (meeting [ACC], [be-in′ (office [ACC], [[do′ (A: PSA
[NOM] Linda, Ø)] CAUSE [INGR see′ (NMR: to [ACC] salesman, U: [ACC] what)]])])]⟩⟩
{ACT: 3sg}
(118) Step 5. Assign the nucleus, arguments and adjuncts to positions in the
syntactic representation of the sentence.
a. Assign the nucleus to the appropriate position in the core.
b. Assign the [WH] arguments to the appropriate positions in the clause.
c. If there is a [þWH] XP,
1. assign it to the normal position of a non-wh-XP with the same function,
or
2. assign it to the pre-core or post-core slot, or
3. assign it to a position within the potential focus domain of the clause
(default ¼ the unmarked focus position).
d. A non-wh-XP may be assigned to the pre-core or post-core slot, subject to
information structure restrictions (optional).
e. Assign the argument(s) of LS(s) other than that of the predicator in the
nucleus to
1. the core- or clause level periphery (default), or
2. the pre-core or post-core slot, or
3. the pre-detached or post-detached position.
The formulations in (118a, b) are intentionally vague and for good reason:
every language does it differently. While there are well-known cross-linguis-
tic tendencies with respect to word (or phrase) order, languages differ in the
very fine-grained constraints on word order, even if they follow these ten-
dencies with respect to larger-grained phenomena. As mentioned earlier, it
is possible to include the nucleus in the English core templates because its
position is virtually always the same: after the PSA and before the other core
arguments and phrasal adjuncts. This is not possible in languages like
German, Croatian and Dyirbal, in which the nucleus has no fixed position,
and in these three languages the linearization constraints on its position in
the core are different. With respect to the linearization of non-wh argument
expressions, the constraints on their order vary dramatically across lan-
guages, as a brief comparison of, for example, English, Croatian, Dyirbal
and German would show.
The situation regarding wh-expressions is considerably simpler: there are
basically three possibilities. First, they could occur in situ, that is, the normal
position of a non-wh-expression with the same function. Second, they could
appear in the PrCS or PoCS, as in English and Dhivehi (see (6)), and third,
they could appear in the unmarked focus position in the clause (see Chap-
ter 11), as in Turkish and Basque. It should be noted that occurrence of a
semantic argument in the PrCS or PoCS does not involve movement, as the
wh-expression is linked directly from the semantic representation to the
PrCS or PoCS; furthermore, there is no co-indexed empty RP or
Figure 1.36 Summary of the linking from semantics to syntax for (109)
d. The information from these steps should link everything in the core to the
argument positions in the LS; if there is an element in the PrCS, it will be
linked last, to the remaining unlinked argument position in the LS. If
there are peripheral adjuncts, retrieve the LS of the predicative adposition
and add this LS to the semantic representation.
The linking presupposes that the parser has output a labelled syntactic
tree. This is an idealization that is appropriate for an abstract grammar, but
it does not reflect the nature of real-time sentence processing, in which
interpretation is incremental and begins with the recognition of the first
word of the sentence and does not wait for the entire sentence to be uttered
before starting the interpretive process. (See Chapter 19 for a proposal which
attempts to account for incremental interpretation in an RRG-based sen-
tence processing system.)
The first step is to glean all information from the morphosyntax of the
sentence. In a language with a voice opposition, the voice of the predicator
in the nucleus is crucial, since it signals the function of the PSA, in English,
the first RP in the core, and in German, the nominative RP. With intransi-
tive predicators, on the other hand, voice and the PSA Selection Hierarchy
are irrelevant, because there is only one direct core argument, which can
be either actor or undergoer, depending on the semantics of the predica-
tor. In addition, there is a wh-RP in the PrCS; it carries no clues as to its
function in the sentence in English, as in principle it could be the PSA (e.g.
What upset Mary?), a non-PSA undergoer (e.g. What did Bill buy?), a non-
macrorole direct core argument (e.g. What did Sally show Linda?), or an
oblique core argument (e.g. What did they load the truck with?). In addition,
there are three PPs, one in the core, one in the core-level periphery, and
one in the PrDP.
The second step is to call up the LS of the predicator in the nucleus, in this
case, show, and assign macroroles. The x-argument is assigned actor, but it is
not possible to assign undergoer, due to the fact that either y or z could be
the undergoer with this verb.
The final step is to reconcile the results of the first two steps in order to
satisfy the Completeness Constraint in (107). There is one direct connection:
the RP Linda is the actor, the actor is the x-argument in the LS of show, and
therefore Linda is the x-argument. The other RP in the core, the salesman, is
the object of the non-predicative preposition to and must be linked to either
the y- or the z-argument in the LS. The key here is the preposition assign-
ment rule in (115), which states ‘Assign to to the non-macrorole a argument
in the LS configuration BECOME/INGR/FIN predicate′ (a, b)’. In terms of the
LS for show, the y-argument is the one corresponding to the ‘non-macrorole a
argument in the LS configuration INGR predicate′ (a, b)’ , and accordingly
the RP the salesman must be linked to the y-argument, as shown in
Figure 1.47. That leaves one unlinked RP argument in the PrCS, what, and
one unlinked argument position in the LS, namely the z-argument. In order
to satisfy the Completeness Constraint, the wh-RP must be linked to the z-
argument, and this yields the correct interpretation of it.
The Completeness Constraint is not yet fully satisfied, however, as there
are two unlinked PPs in the sentence. These are adjunct predicative PPs
which license their arguments and must be incorporated into the semantic
representation of the sentence. Since the prepositions are predicative, they
have lexical entries in the lexicon, and accordingly it is necessary to go back
to the lexicon and retrieve their LSs. The question arises as to the order of
composition of the two prepositional predicates. The locative PP is inside the
clause in the core-level periphery, while the temporal PP is a frame-setting
topic outside of the clause in the PrDP. Consequently, the locative LS is
added, taking the LS of show as one of its arguments, and then the temporal
LS is added, taking the locative LS as one of its arguments. When the
operators are added (not shown), the result is the semantic representation
in (116).
SYNTAX:
PSA: (111b1, c1); Variable [±pragmatic influence]
Linking: PSA-modulation, Argument-modulation (core-level peripheral PP, Ø)
Template(s): (117b2)
MORPHOLOGY:
Verb: Past participle
AUX: werden (PERF: worden)
P: von
SEMANTICS:
PSA is not instigator of state of affairs but is most affected participant (default)
PRAGMATICS:
Illocutionary force: Not imperative
Focus structure: Predicate focus (default)
Figure 1.37 Constructional schema for German ‘plain’ passive57
quite different properties from the ‘plain’ werden-passive (Van Valin 2003;
Diedrichsen 2004).
The schema specifies the defining syntactic, morphological, semantic and
pragmatic properties of the construction.58 The first line under ‘syntax’,
‘PSA’, states that it is an accusative construction by referring to the PSA
Selection Hierarchy in (111b1) and that the PSA in this construction must be
a macrorole argument, (111c1). ‘Variable [pragmatic influence]’ indicates
that the PSA is not fixed and that the choice of which macrorole functions as
PSA may be influenced by information-structure factors (see LaPolla (Chap-
ter 5), Bentley (Chapter 11), Latrouite and Van Valin (Chapter 12)). The next
line identifies the construction as a voice construction involving both PSA-
and argument-modulation and specifies that argument modulation is
instantiated either by the default choice for PSA occurring in a PP in the
core-level periphery or by it being omitted. This is indicated by inserting a
‘Ø’ in the leftmost argument position of the LS. The third line states that the
exception in (117b2) for argument-modulation voice construction comes
into play in the selection for the syntactic template. Thus, the entries under
‘syntax’ in the CS give the language-specific values of the cross-linguistically
valid principles in the theory.
The next entries under the heading ‘morphology’ give the relevant morpho-
logical properties of the construction. The first entry gives the required form
of the verb, namely the past participle, and the second identifies the auxiliary
used in the construction, namely werden ‘become’, which has an irregular
past-participle form when it is used as a passive auxiliary, worden, which lacks
the normal ge- prefix: Es ist wärmer geworden ‘It has got warmer’ (*Es ist wärmer
worden) vs. *Der Rechner ist von Hans kaputtgemacht geworden (cf. (120b)). The final
part specifies the P in the adjunct PP which contains the actor, the default
choice for PSA. It is von ‘from, of’, and the fact that it takes a dative object
need not be specified here, as that is a general property of von.
The characterization of the meaning of the construction under the
‘semantics’ heading is quite general but appropriate for the construction.
There are two entries under ‘pragmatics’. The first states that this construc-
tion can be used in assertions and questions but not in imperatives, and the
second reflects the fact that the default focus structure is predicate focus,
with a topical PSA and a focal predicate phrase (see Chapter 11), and
accordingly in a passive construction, just as in an active voice form, the
nominative PSA is interpreted as the default topic, although it must be
emphasized that this is only the default interpretation.
CSs thus represent language-specific information, formulated in terms of
the appropriate general, cross-linguistically valid principles. The schemas
play an important role in the linking between form and meaning. In Van
Valin (2005) there is a detailed discussion of the integration of the linking
algorithm with CSs in Sama wh-questions (pp. 132–135, 157). In order to
illustrate the interaction of CSs with the linking algorithm, the schema for
the Sama antipassive is taken from Van Valin (2005) and presented here in
SYNTAX:
PSA: (111b2, c1); Variable [±pragmatic influence]
Linking: PSA-modulation, argument-modulation: optional if 3p (Ø), obligatory if
~3p] (Ø, core PP)
Template(s): (117b) [only with Ø argument-modulation]
MORPHOLOGY:
Verb: N- + stem
P: ma
SEMANTICS:
PSA is instigator of state of affairs
If predicate is stative, then it shifts to activity in actor voice
If lowest ranking argument is not marked by determiner, then preferentially indefinite
or non-referential
PRAGMATICS:
Illocutionary force: Unconstrained
Focus structure: Predicate focus (default)
Figure 1.38 is an updated form, along with the relevant data in (121)
(Walton 1986, personal communication). The homorganic nasal prefix N-
which marks actor voice is realized as Nga- when prefixed to a verb begin-
ning with /n/.
The relevant facts about actor and undergoer voices are summarized in
the pairs of examples in (121a, b); the first example is undergoer voice
(active) and the second is actor voice (antipassive). In the (a, a′) examples
the actor is either first person (aku/ku) or third person (d’nda ‘woman’), and
the undergoer is third person (onde’ ‘child’). Sama has case marking only on
pronouns. In the undergoer (active) voice the full RPs are distinguished by
word order, while the pronominal actor occurs in the ergative form. In the
actor (antipassive) voice, the pronominal actor occurs in the nominative
form, signalling that it is the PSA. A striking feature of this construction is
that when the undergoer is third person, it may occur as a direct core
argument.59 If, however, the actor is third person and the undergoer is first
or second person, as in the (b, b′) examples, the situation is very different. In
the undergoer voice, the first-person undergoer occurs in the nominative
form aku, as in (121b), but in the actor voice the first-person undergoer is
marked by the oblique preposition ma plus the nominative pronoun, as in
(121b′). The CS for this voice construction is given in Figure 1.38.
The PSA entry specifies that this is an ergative PSA restricted to macro-
roles, while the linking entry states that the primary function of the con-
struction is PSA-modulation, with argument-modulation optional for third-
person arguments (omission), but obligatory for non-third-person argu-
ments, the options being omission or occurring as an oblique core argument
in a PP as in (121b′). The final entry states that the default syntactic template
selection principle in (117a) applies unless there is optional argument-
modulation, and then only when the choice is omission, since the PP for
non-third persons is in the core, not a periphery. The relevance of this CS to
wh-questions in Sama is clearly illustrated in (121c, c′). Wh-expressions like
say ‘who’ or ay ‘what’ occur in the PrCS and must be the PSA of the clause.
Accordingly, if the verb is in undergoer voice, as in (121c), the wh-expression
must be interpreted as the undergoer, and d’nda ‘woman’ must be inter-
preted as the actor. On the other hand, if the verb is in actor (antipassive)
voice, then the wh-expression is the actor and the postverbal RP d’nda’ must
be interpreted as the undergoer. The CS for wh-questions invokes the anti-
passive CS when the wh-expression is an actor.
Thus, the schema for the Sama antipassive, like the one for the German
passive, combines cross-linguistically motivated theoretical principles with
language-specific information. The interaction of CSs with the linking algo-
rithm will be discussed in more detail in the next section. The issue to be
investigated in the remainder of this section is the range of ways that
language specific-grammatical information can be represented.
CSs are one of the primary devices for representing language-specific gram-
matical information. Some cases of word-order variation lend themselves
readily to a constructional analysis, for instance the occurrence of a phrase
in an extra-core position (PrDP, PrCS, PoCS, PoDP) subject to information-
structural constraints. Some basic word-order patterns also seem amenable
to a constructional analysis, even though a given word order need not be
SYNTAX:
PSA: (111b1, c2); Variable [±pragmatic influence]
Linking: PSA-modulation, argument-modulation: (Ø, core-peripheral PP)
Template(s): (117b)
MORPHOLOGY:
Verb: Stem + (r)are
P: ni(yotte)
SEMANTICS:
PSA is not the instigator of state of affairs but undergoes the action of the verb.
PRAGMATICS:
Illocutionary force: Not imperative
Focus structure: Predicate focus (default)
Figure 1.39 Constructional schema for Japanese plain passive
SYNTAX:
PSA: (111b1, c2)
Linking: Highest ranking argument of stem verb —> NMR
Template(s): (117a)
MORPHOLOGY:
Verb: Stem [renyookee ‘infinitive’ form] + (r)are
Case: Assignment follows Imai (1998)
SEMANTICS:
-(r)are adds feel.affected′ (x, [y]) [MR Ø] to LS of stem verb y
PSA feels negatively affected by state of affairs
PRAGMATICS:
Illocutionary force: Not imperative
Focus structure: Predicate focus (default)
Figure 1.40 Constructional schema for Japanese adversative passive
to the argument licensed by -(r)are, since, with a transitive verb like taberu
‘eat’ in (125), the accusative RP is an undergoer.62 Because the non-macrorole
argument introduced by -(r)are is the PSA, it does not take dative case,
following the more complex set of case assignment rules proposed for
Japanese in Imai (1998) instead of the ones given in (114a).
The steps in which the CS in Figure 1.39 plays a role in the linking from
semantics to syntax for (122b) are summarized in (126); the steps were given
in (108).
As noted in Section 1.6.1, steps 1 and 2 take place in the lexicon, while
steps 3–5 take place in the syntax, outside of the lexicon. In the plain passive
in (122), both active and passive versions of the sentence have the same LS,
given in (122c), and accordingly it does not involve steps 1 and 2. By contrast,
steps 1 and 2 are crucial for the adversative construction, as the combining
of the LSs takes place at step 1 and macrorole assignment (or the lack
thereof ) at step 2. The remaining three steps in (127) reflect the conse-
quences of steps 1 and 2. In (126), on the other hand, all of the relevant
steps are outside of the lexicon, namely, steps 3 through 5. The adversative
construction is, therefore, considered to be a lexical construction, while
the plain passive is a syntactic construction. This distinction has
important consequences for certain phenomena, for example reflexive bind-
ing (see Van Valin 2005, §5.2).
SYNTAX:
PSA: (111b1, c2)
Linking: Passive construction (optional), variable undergoer selection (default)
Template(s): (117b1,2); RP, PP templates, see CSs for RP, PP
Order: [S [CL [C RPPSA {AUX …}[N PRED ] (RPU) (RP/PP)] Adjunct(s)]]
MORPHOLOGY:
Auxiliary: [±past] PSA agreement; [-tense] bare infinitive, participle[±past]
Verb: [±past] PSA agreement; [-tense] bare infinitive, participle[±past]
RP Case: [+PRO] NOM, ACC, GEN; [-PRO] GEN
SEMANTICS:
LS of PRED in nucleus plus arguments and adjuncts
PRAGMATICS:
Illocutionary force: Unspecified
Focus structure: Predicate focus (default)
Figure 1.41 Constructional schema for basic clause structure in English
SYNTAX:
PSA: none
Linking [semantics to syntax]: WH-XP in PrCS
[syntax to semantics]:
Arguments: After all of the XPs in the core have been linked, link the
WH-XP to the remaining unlinked argument position in the semantic
representation of the core.
Adjuncts: If there are no unlinked argument positions in the semantic
representation of the core, treat the WH-word like a predicative
preposition and follow the CS for linking adjuncts
Template(s): (117b3), PrCS
MORPHOLOGY:
X[+tense] must be the first element in the core linearly.
SEMANTICS:
Contains an open proposition with variable α , WH-XP = α.
PRAGMATICS:
Illocutionary force: Interrogative
Focus structure: Narrow focus on PrCS
Meredith in the hospital, be-in′ (hospital, [do′ (Sam, [visit′ (Sam, Meredith)])]), or
Sam visited Meredith after the party, be-after′ (party, [do′ (Sam, [visit′ (Sam,
Meredith)])]). In both examples the first argument is the spatial or temporal
reference point and the second argument is the event or state of affairs being
located in space or time. While they all have a default periphery in which
they occur, they are also subject to displacement and occurrence in the PrCS
and PrDP. In addition, there can be multiple adjuncts, and their ‘scope’-like
relations are constrained by both semantic and pragmatic factors. Many of
the aspects of the CS in Figure 1.43 would be applicable to other languages;
some would require only a superficial reformulation in terms of word order,
for example a language which has postpositions rather than prepositions,
but some would be quite substantial, for example languages which lack a
PrCS and/or a PrDP, and, more significantly, languages such as Dyirbal that
lack adpositions altogether. In languages with rich case systems, very often it
is cases which mark adjuncts rather than adpositions (e.g. Hungarian), and so
it is necessary to recognize that there can be predicative cases as well as
predicative adpositions (see Nakamura 2021 for a discussion of predicative
cases in Russian).
It is now possible to formulate the syntax-to-semantics linking algorithm
in such a way that it contains only cross-linguistically valid steps, the
language-specific information being contained in a language’s CSs. The goal
here is to redo the linking in Section 1.6.2 based on information supplied by
the CSs in Figures 1.41–1.43.
SYNTAX:
PSA: none
Linking [semantics to syntax]:
Assign the argument of an LS other than that of the predicator in the
nucleus to a predicative PP headed by the predicative P, or identify a
one-place adverbial predicate, and instantiate them in
1. a periphery [(81i) = core, (81j) = clause] (default), or
2. the PrCS or the PrDP, subject to information-structural conditions (see
PRAGMATICS).
[syntax to semantics]:
If there are no unlinked argument positions in the semantic represen-
tation of the clause, retrieve the LS for the predicative preposition or
he adverbial from the lexicon; if the predicative element is a one-place
adverb, insert the LS of the core into its argument position, and if it
is a two-place preposition, insert the LS of the core into the second
argument position and insert the object of the P as the first argument.
Template(s): periphery, predicative PP, adverb, PrDP, PrCS
MORPHOLOGY: none
SEMANTICS:
Semantic relations: (81i, j)/(94i, j)
When there are multiple adjuncts, the order of linking reflects (1) clausal
periphery > core periphery, (2) PrDP > PrCS > default periphery, or (3) [PER
AJTTOP > AJTFOC ], where the lower ranking adjunct is linked first.
PRAGMATICS:
Illocutionary force: Unspecified
Focus structure:
Frame-setting topic in PrDP
Contrastive focus/topic in PrCS
Neutral comment, narrow focus in periphery
Among its many functions the verb prefix be- serves to signal a non-default
choice for the argument to function as undergoer. Most languages which
have variable undergoer selection are not so straightforward. Rather, the
identity of the argument functioning as undergoer must be deduced from
the way the non-macrorole argument is coded. The following English
examples illustrate this.
Figure 1.48 The linking from syntax to semantics in terms of the algorithm in (128)
The issue here is clear: both non-actor arguments are in the same case in the
active voice, and therefore the principles in Steps 4a, b fail to distinguish
between the two arguments in the linking.
The Icelandic verb skila ‘give back, return’ is a three-place predicate whose
non-actor arguments exhibit two different patterns: two dative RPs (recipi-
ent, theme), as in (131a), and a dative RP (theme) and a PP (recipient), as in
(131c). The latter follows the rule in Step 4a, since linking the object of the
preposition til ‘to’ to the y-argument in the LS yields the correct result. This
requires that the preposition til ‘to’ outweigh the dative case on the theme
argument in Step 4a, and consequently this language-specific feature must
be expressed in a CS concerning linking with three-place predicates. The
passive in (131c′) illustrates the fact that the PSA in Icelandic does not have
to be a macrorole; peningunum ‘the money’ has the syntactic privileges of a
nominative PSA, despite not having the coding privileges of one (i.e. nom-
inative case, ability to be an agreement trigger) (see Van Valin 1991).
The pattern with two dative RPs in (131a) renders Steps 4a, b inapplicable,
since the two RPs bear the same case. There is, nevertheless, an important
difference between the two direct core arguments: one is human and the
other is inanimate, and this difference correlates with the way the RPs link
to the LS. The human RP is the recipient (y), while the inanimate RP is the
theme (z). This can be captured in the following step in the Icelandic CS for
three-place predicates.
When the two non-actor RPs in the core are both in the same case, link the
human (animate) RP with the first argument position in the two-place state
predicate in the LS.
This linking rule comes into play in Icelandic only when there are two dative
RP objects in the core; otherwise steps 4a, b in (128) are operative.
A comparison of the Yaqui examples in (132) with the Icelandic examples
in (131a–c) suggests that a principle like (133) is part of the Yaqui linking
algorithm as well, with double-accusative RP objects instead of two dative
RPs. The Yaqui situation is more complex, however, due to sentences involv-
ing valence-increasing constructions, such as derived causative verbs, as
illustrated in (134).
These examples have three core arguments in the accusative case, two of
which are human, and only one can be interpreted as the causee and only
one as the recipient. The sentences in (134a, b) are unambiguous: in (a) usi-
‘the child’ can only be interpreted as the causee and yoem- ‘the man’ as the
recipient, and vice versa in (b). Accordingly, the correct linking in this
construction cannot refer simply to case or to animacy.
It was noted in fn. 52 that Yaqui verbs show three different undergoer
linking patterns, termed patterns A, B and C, following Guerrero and Van
Valin (2004). Pattern A is the usual direct–indirect object pattern, exem-
plified by nenka ‘sell’ in (135a), in which the lowest-ranking argument in
the AUH is the undergoer, is in the accusative case, and can be the PSA in
a passive construction, as in (135a′). Pattern B is the primary object
pattern (Dryer 1986), illustrated by miika ‘give’ in (132), in which the
undergoer is the second highest-ranking argument in the AUH, all non-
actor core arguments are in the accusative case, the undergoer normally
immediately follows the nominative actor, and it alone can be the PSA in
a passive. Pattern B not only applies to miika ‘give’ and the other verbs in
this class, but also to derived forms involving valence-increasing construc-
tions, namely causative, applicative, desiderative, reported speech and
propositional attitude. (See Guerrero and Van Valin 2004 for exemplifi-
cation.) Pattern C is defined by variable undergoer selection, alternating
between Patterns A and B, as shown in (135b, c) with chijakta ‘sprinkle’.
Actor and undergoer selection are best handled in terms of a CS, since
undergoer selection has several language-specific facets. In terms of seman-
tics-to-syntax linking, step 2 in (108b) is the relevant step, and pattern
A follows the default pattern of the AUH. Pattern B, on the other hand, does
not and specifies that the undergoer is the second-highest-ranking argument
in the LS. In a two-argument LS (e.g. jamta ‘break’ [do′ (x, [act.on′ (x, y)])]
CAUSE [INGR broken′ (y)]), ‘lowest ranking’ and ‘second highest raking’ pick
out the same argument, namely y, but in a three-argument LS like miika
‘give’ ([do′ (x, Ø)] CAUSE [INGR have′ (y, z)]) they select different arguments as
undergoer, z in pattern A and y in pattern B. The verbs which exhibit just
pattern A or B do not alternate with respect to undergoer selection, and they
each occur with different case marking. Pattern A verbs like nenka ‘sell’ in
(135a) have an accusative undergoer/theme, (z), while the recipient (y) carries
the postposition -u, making it an oblique core argument. Pattern B verbs like
miika ‘give’ take all of their non-actor direct core arguments in the accusa-
tive case. The undergoer is distinguished by word order: it normally imme-
diately follows the nominative actor, and if it is not immediately after the
NOM RP, it is the first one after it satisfying the selectional restrictions of the
verb. Pattern C verbs like chijakta ‘sprinkle’ show the same case alternation
found in many other languages, namely ACC (U) – LOC (locative oblique core
argument) vs. ACC (U) – INS (instrumental oblique core argument), as in
(135b, c). In all three patterns the undergoer appears in the accusative case;
it is the case marking of the other core arguments that varies across the
three classes of verbs.
The Animacy Principle was introduced in (133) to account for the Icelandic
examples in (131), and based on the superficial similarity between the sen-
tences in (132) and (131) it was suggested that it might be operative in Yaqui,
too. It could be argued that it applies to (132a), for example, but this is also
readily accounted for by Step 4 in (128). More problematic for the Animacy
Principle are the sentences in (136), from Guerrero and Van Valin (2004).
In (136a) there are three human RPs, and the sentence is actually ambigu-
ous. It could be argued that the Animacy Principle correctly predicts the
ambiguity of (136a), since it makes it possible for either the RP Karmen or the
RP usi ‘child’ to be interpreted as the first argument in the two-place state
predicate in the LS. With respect to four-place derived predicators, as in
(134), they are not ambiguous, and this can be accounted for by having the
Animacy Principle apply after the actor and undergoer have been identified.
Accordingly, in (134a), after maejto ‘teacher’ is determined to be the actor
and usi- ‘child’ the undergoer, the Animacy Principle can apply to the
remaining two core arguments, and the result is that yoem- ‘person’ is
correctly analysed as the recipient/experiencer, and mansana ‘apple’ is cor-
rectly analysed as the thing given (theme) in the central LS of the events
described in (134).
The sentence in (136b), however, cannot be accounted for by (133), because
all of the non-NOM RPs are inanimate. Hence (133) does not apply, and this is
also the case when the sentence is causativized, as in (137).
After determining that Goyo is the causer, actor and PSA and Tibu is the
causee and undergoer, there are still two inanimate arguments left to be
accounted for. All verbs have selectional restrictions constraining the type of
RPs that can function as their arguments, and in RRG they are expressed as
qualia properties (Pustejovsky 1995), both of the verb and of the RPs, such
that the qualia properties of the RPs must be compatible with the selectional
restrictions of the verb’s arguments. This can be formulated as the ‘Qualia
Principle’ as in (138).
In (136b) and (137), the crucial part of the LS is ‘CAUSE [INGR NOT be-LOC ′
(y, z)]’, where y is the ‘container’ and z are the things in it. With respect
to rifles and bullets, it is clear that rifles can have bullets in them, not
the other way around, and therefore wikoi- ‘rifle’ is the y-argument and
juiwa- ‘bullet’ is the z-argument. Thus the interpretation of (136b) and
(137) derives from the very specific qualia properties of these RPs. By
contrast, invoking the Animacy Principle in (133) to account for (131a)
involves only very general qualia properties, namely animacy. Thus, the
Animacy Principle can be subsumed under the Qualia Principle.
The difference in complexity between what is required for the semantics-
to-syntax linking and the syntax-to-semantics linking is striking. The
former is summarized in (139) and the latter in (140). The steps referred to
in (139) are those in (108), whereas the steps in (140) are those found in
(128). With respect to (139), it is important to keep in mind that patterns
A and B do not involve variable undergoer selection; that is solely a
property of pattern C.
The point of this discussion of Icelandic and Yaqui is to show what kinds
of principles, both language-specific and cross-linguistically valid, are
needed. The Qualia Principle is a very general principle governing the
relationship between the core arguments and the predicator in verbal cores,
whereas the specification that all non-PSA direct core arguments with pat-
tern B verbs must have accusative case is clearly idiosyncratic to Yaqui. The
information in (138) and (139) would be stored in CSs, under the heading of
‘Linking’ in the syntax section of the CSs.
(118) Step 5. Assign the nucleus, arguments and adjuncts to positions in the
syntactic representation of the sentence.
a. Assign the nucleus to the appropriate position in the core.
b. Assign the [WH] arguments to the appropriate positions in the clause.
c. If there is a [þWH] XP,
1. assign it to the normal position of a non-wh-XP with the same function, or
2. assign it to the pre-core or post-core slot, or
3. assign it to a position within the potential focus domain of the clause
(default ¼ the unmarked focus position).
d. A non-wh-XP may be assigned to the pre-core or post-core slot, subject to
information structure restrictions (optional).
e. Assign the argument(s) of LS(s) other than that of the predicator in the
nucleus to
1. the core- or clause level periphery (default), or
2. the pre-core or post-core slot, or
3. the pre-detached or post-detached position.
(141) Step 5: Assign the nucleus, arguments and adjuncts to positions in the
syntactic representation of the sentence.
a. See CS for English basic clause structure [Figure 1.41] (¼a, b)
b. See CS for English displaced wh-questions [Figure 1.42] (¼c)
c. A non-wh-XP may be assigned to the pre-core or post-core slot, subject to
information structure restrictions (optional). [a yet to be formulated CS]
d. See CS for English peripheral adjuncts [Figure 1.43] (¼e)
This better represents what English speakers know about their language and
yet still maintains a balance between cross-linguistically valid linguistic
concepts and principles and the language-specific data that speakers rely
on when they actually use language.
actors are responsible for the state of affairs, in the sense that in order to
have a perceptual event there must be a perceiver, for a cognition state of
affairs a cognizer, for an eating event an eater, etc., while undergoer is
associated with affectedness. It appears that languages tend strongly to
lexicalize responsibility but not affectedness, leaving the choice of which
participant is most affected to the speaker in some languages. Languages
vary with respect to whether they allow variable undergoer selection or not.
If they do, there is variation in the number of verbs allowing the variable
selection, and there are differences in terms of the definition of undergoer
in terms of the AUH. As shown earlier, Yaqui has a class of verbs for which
the undergoer is the lowest-ranking argument in the LS of verbs and another
class in which it is the second-highest-ranking argument.
Step 3 involves PSA selection and case marking, and here the cross-
linguistic variation dwarfs that found in steps 1 and 2. The variation in
PSA selection includes syntactically accusative languages without a voice
system, syntactically accusative languages with a productive voice system,
syntactically ergative languages with a productive voice system, mixed
systems with first and second person being syntactically accusative and
third person being syntactically ergative, inverse systems, split-intransitive
systems, Philippine-type languages with multiple voice systems, which can
be syntactically accusative, ergative or symmetrical, and systems which
defy categorization in terms of traditional categories (e.g. Barai, Papua-
New Guinea; Van Valin 2009). Case-marking systems are likewise highly
varied, and they don’t always match the syntactic pattern, for instance it
has long been known that most morphologically ergative languages lack
ergative syntax.
Step 4 is concerned with the selection of the syntactic templates that
constitute the syntactic representation of the sentence, and here too there
is great diversity, given that the templates must embody the syntactic
structures of the full range of human languages. The components of the
LSC fall into two basic groups: those that are semantically motivated (see
Section 1.2.2) and are universal (nucleus, core, clause, peripheries) and
those that are not semantically motivated but rather are pragmatically
motivated and are not universal (PrDP, PrCS, PoCS, PoDP). Step 5, as shown
earlier, incorporates language-specific CSs capturing the full sweep of
linearization options, displacement options, placement of adjuncts, etc.
Thus, the post-lexical steps in the linking algorithm exhibit much greater
cross-linguistic variation than the lexical steps, and this is indicative of
the inverse relationship between semantic motivation and cross-linguistic
variation.
Two further examples will be presented to illustrate this point. The first
concerns the LSC. There are obligatory and optional constituents in it. The
obligatory ones are the nucleus, the core, the clause and the sentence, and
within the core are the arguments of the predicator in the nucleus. As
argued in Section 1.2.2, the fundamental distinctions in the LSC follow from
(142) a. Max tried to tune the piano. Max is the actor of try and the actor of tune.
b. Sally persuaded Max to tune Max is the undergoer of persuade and the
the piano. actor of tune.
c. Sally promised Max to tune Sally is the actor of promise and of tune.
the piano.
d. Sam brought wine for Wine is the undergoer of bring and drink.
everyone to drink.
(143) a. Sam seems to really like wine. Sam is a syntactic core argument (and
PSA) of the core headed by seem but is a
semantic argument of like.
b. Sally believes Sam to really Sam is a syntactic core argument of the
like wine. core headed by believe but is a semantic
argument of like.
In (142) each clause contains two cores, and there is a single RP which is a
semantic argument of both predicators in the nucleus of each core. In
(143), on the other hand, there is no single RP which is a semantic argu-
ment of each predicate; rather, there is an RP which is a syntactic core
argument, but not a semantic argument, of one core, and a semantic
argument of the predicator in the linked core. Thus, there are two distinct
ways for two cores to have an argument in common: the constructions in
(142) involve what is traditionally known as (obligatory) control, while
those in (143) are traditionally known as raising.65 The argument in
common is technically known as the shared argument in the
construction.
In an obligatory control construction, a participant functions as a seman-
tic argument in each core but is instantiated in only one of the cores. In
sentences like (142a) there is only one core argument in the licensing core,
and it must be the shared argument. In (142b, c) there are two core
arguments in the licensing core, and there has been much discussion over
the past almost sixty years regarding the explanation for the difference
between verbs like persuade (so-called ‘object control’) and promise (so-called
‘subject control’). The solution was given in Foley and Van Valin (1984:
306–311), and it was shown in Van Valin and LaPolla (1997, §9.1.3.1) and
Van Valin (2005, §7.3.1) to be valid for a typologically wide range of
languages. It is semantic: if the licensing verb has causative or jussive
semantics, the undergoer is always the controller, whereas if it does not
have causative or jussive semantics, the actor is the controller. Because this
principle is semantic, it would be expected in light of the discussion in
Section 1.6.5 that it would be universally valid, and as far as is known, this
is the case. This semantic account makes a prediction that syntactic
accounts do not: if a verb can have either jussive or non-jussive semantics,
the choice of controller should change with the semantic change in the
verb. This is illustrated in (144).
(144) a. The teacher asked the rowdy student to leave the classroom.
b. The sick student asked the teacher (to be permitted) to go home.
In (144a) the verb ask has jussive semantics, and the undergoer is the
controller, as predicted. In (144b), on the other hand, it does not have jussive
semantics, as it is a request, and the actor is the controller, as predicted.
Because there is a shared argument in these constructions, the linked core
has one core argument less than if it were a core in a simple sentence, and
therefore the syntactic template selection principles in (117) must be modi-
fied. Unlike the language-specific principles in (117b2, 3), this one is univer-
sal. There is one additional principle affecting non-subordinate core
junctures, which is illustrated here.
Section 1.3.2 it was shown that the infinitival cores in these constructions
are not embedded, and therefore they are not core arguments in the licens-
ing core. Accordingly, the core containing the licensing predicate in a non-
subordinate core juncture has one less core argument than when it is a
simple core, as the contrast between (145a) and (a′) shows. The same holds
for two-place predicates like forget in (145b, b′): when it heads a simple core,
as in (145b), it has two core arguments, but when it is the head of the
licensing core in a non-subordinate core juncture, as in (145b′), it has only
one. Thus, the verb in the licensing core in a non-subordinate core juncture
has one less core argument than when it is the head of a simple core with RP
or PP arguments only.
The LSs for (144a′, b′) are given in (146).
(146) a. [do′ (Bill, Ø)] CAUSE [BECOME know′ (Maryi, [do′ (zi, [swim′ (z)])]
b. forget′ (Sami, [[do′ (yi, Ø)] CAUSE [INGR locked′ (door)])])
In (148a) the controller and shared argument uži- ‘boy’ is in the licens-
ing core, which is shown by the fact that it carries dative case assigned by
the predicator šuλ’i- ‘forget’, and it is co-indexed with the lexically
unfilled variable in the embedded LS corresponding to the linked core.
This example adheres to (147b1, 2). In (148b), on the other hand, the
situation is very different, as the shared argument and controller is in
the linked core, not the licensing core, as shown by the fact that it carries
ergative case, being the actor of the transitive verb ‘feed’, rather than
absolutive case as the single core argument of the intransitive verb -oq-
‘begin’. As the LS shows, the lexically unfilled variable is in the LS of the
licensing core, and the lexically filled argument is in the LS of the linked
core. Both constructions satisfy the Completeness Constraint, and there is
nothing in the theory which precludes the two possibilities in (148). The
question of why the one option is so much more frequent than the
other remains.
An interesting feature of the (b) example is that the controller in the
linked core triggers noun class agreement on the verb in the licensing core.
This is readily accounted for: ‘begin’ shows noun class agreement with the x-
argument, and the x-argument is co-indexed with kid ‘girl’ (class II) in the
embedded LS, and accordingly it will show class II agreement. The structure
of (148b) is given in Figure 1.49.
This structure violates (147b2), because the linked core has its full comple-
ment of RP arguments, unlike its English translation. It could also be
construed as violating (147b1), because it lacks the core argument in the
licensing core that the principle predicts it should have: -oq- ‘begin’ has two
arguments in its LS, and when it occurs in the licensing core of the juncture,
it should have one core argument.
which has three components: dǎ ¼ [do′ (1sg, [beat′ (1sg, shé)])], sı̌ ¼ [INGR
dead′ (shé)], and cause, which represents the constructionally invoked
meaning of causation. The derived LS is very similar to that of a lexical
causative verb, for example smash as in ‘The burglar smashed the
window’ ¼ [do′ (burglar, [act.on′ (burglar, window)])] CAUSE [INGR
smashed′ (window)]. Nuclear junctures with the second predicate being
transitive link like lexical ditransitive verbs, as in the French example in
(149a) (Hyman and Zimmer 1976).
Macrorole assignment works the same with reference to both LSs, yielding
the prototypical ditransitive linking pattern for both.
The cores in non-subordinate core junctures necessarily ‘share’ a core
argument, as shown in the examples in (142) and (143). The shared argument
was termed the ‘argument in common’ earlier. The constituent projections
of (142a, b) are given in Figures 1.50 and 1.51. (See Section 1.3.2.)
If persuade in Figure 1.51 were replaced by promise, as in (142c), the struc-
ture would not change, because the difference between actor control and
undergoer control is captured by the theory of obligatory control as part of
the linking algorithm. The important point for the discussion of argument
sharing is that there is no empty argument position or zero element in the
linked core. The essential features of the linking from semantics to syntax
are given in (150).
(150) do′ (Maxi, [try′ (Max, [[do′ (xi, Ø)] CAUSE [INGR tuned′ (piano)]])])
Actori Actori Undergoer
PSA DCA
(151) a. Sam doesn’t/didn’t seem (to Sally) to like (any of ) the wines.
b. Sam doesn’t/didn’t seem (to Sally) to be eating any cookies.
c. Sam doesn’t/didn’t seem (to Sally) to have saved a red cent.
Second, Bohnemeyer and Van Valin (2017) argue that core cosubordination
constructions have the macro-event property and therefore they cannot take
(153) a. *This morning Sam didn’t seem to be eating pizza in the afternoon.
b. This morning Sally believed Sam to be eating pizza for lunch at noon.
realized as a direct core argument. From this LS two different patterns can
come, depending on whether there is a tense operator in the embedded LS. If
there is one, the result is It seems to Sally that Sam likes wine. If there is only one
tense operator over the entire LS, the result is (143a), in which the PSA of the
linked core appears as the PSA of the licensing core. This is a function of the
principles in (147b2) and (147c1) together with the M-atransitive nature of
seem: (147b2) requires that the linked core is missing a core argument
position, and (147c1) requires that the licensing core have at lease one core
argument position, which cannot be filled by one of the arguments of seem.
In order to avoid a Completeness Constraint violation, the ‘orphaned’ argu-
ment of like is linked to the open core argument slot in the licensing core,
yielding (143a). This is a different kind of argument sharing from that in
control constructions. It is a very common construction in the languages of
the world.
The ‘raising to object’ construction, on the other hand, is not. The LS for
(143b) would be believe′ (Sally, [like′ (Sam, wine)]), and as before, there are
two possible realizations, depending on whether there are two tense oper-
ators or just one over the entire LS. The first possibility yields Sally believes
that Sam likes wine, and the second yields (143b), which violates the
principle in (147b1). Believe and other ‘raising to object’ verbs are excep-
tional in not reducing their S-transitivity as required by (147b1), thereby
leaving an open core argument slot which can be filled by the ‘orphaned’
argument of the linked core verb, in this case like, thereby avoiding a
Completeness Constraint violation. The nature of argument sharing in this
construction is the same as that in the seem construction. (See also Van
Valin 2005, §7.3.2.)
The purpose construction in (142d), Sam brought wine for everyone to drink,
has an obligatory shared argument, unlike the rationale construction in
(154), which is a clausal juncture.
This construction does not allow a shared ‘object’ argument in the linked
clause, but does permit an optional omitted PSA coreferential with the PSA
in the first clause. The structure of (142d) is given in Figure 1.54; it is core
coordination, as shown by the fact that each core can take an independent
temporal positional modifier, for example Sam brought wine in the afternoon
for everyone to drink at the party during the evening.
The LS for this sentence is [do′ (Sam, [bring′ (Sam, winei)])] PURP [do′
(everyone, [drink′ (everyone, xi)])].68 This construction is significant, because
it involves a controller–pivot relationship in which neither the controller
nor the pivot is a traditional subject (see §1.3.2, Van Valin 2009). Because the
dependent LS is not an argument of the matrix LS, the reduction in
S-transitivity in the licensing core mandated by (147b1) does not apply.
(155) a. Sallyi was talking to Samj, and shei/hej waved to Margaret as she walked by.
b. Sallyi was talking to Samj, and proi/*j waved to Margaret as she walked by.
c. Sallyi was talking to Sam, and waved to Margaret as she walked by.
The first example involves coreference between overt pronouns, and one of
the major constraints in English is the requirement that third-person
pronouns must agree in gender in order to be interpreted as coreferential.
The second example lacks an overt pronoun in the PSA position in the
second clause, and most theories would posit null pronominal, pro, as the
PSA of the second clause. In (155b) only Sally can be interpreted as the one
who waved to Margaret; Sam is ruled out by a syntactic constraint on
coreference to the effect that in conjunction reduction constructions, the
PSA of the first clause must be the controller of the pivot (missing PSA) in
the second clause. It makes sense to talk of coreference between the
controller, RP Sally, and the phonologically null pro as pivot. But RRG does
not permit phonologically null lexical items in its syntactic representa-
tions, and accordingly (155b) is not a possible RRG analysis.69 Rather, the
structure RRG posits is (155c), with no null pronominal. This immediately
raises the question, what is Sally coreferential with? If there’s nothing in
the syntax instantiating the PSA of the second clause, how can this be a
case of coreference? Nevertheless, the PSA of the second clause is inter-
preted to be the same as the PSA of the first clause. The solution cannot be
formulated over the constituent projection of the LSC alone but must
include discourse. In Chapters 11 (Bentley) and 12 (Latrouite and Van Valin)
the RRG approach to the representation of discourse will be presented, and
it is essential to the analysis of conjunction reduction in particular and for
controller–pivot relations in clausal junctures in general. See Section 12.4
in Chapter 12 for detailed presentation of the RRG analysis of conjunction
reduction.
Controller–pivot relations in clausal junctures need not always be gram-
matically constrained, as they are in English, Icelandic, German, and Cro-
atian, for example. Archi, a Caucasian language (Kibrik 1979), lacks syntactic
or semantic constraints on controller–pivot coreference in clausal junctures.
The following examples from Kibrik (1979) illustrate this. In the structural
schemas, the italicized expressions with strikethrough do not occur overtly
in the sentence.
References
Cain, Bruce and James Gair. 2000. Dhivehi (Maldivian). Languages of the
World/Materials 63. Munich: Lincom Europa.
Chief, Liangchen. 2007. Scalarity and Incomplete Event Descriptions in Mandarin
Chinese. PhD dissertation, University at Buffalo (SUNY).
Chomsky, Noam. 1957. Syntactic Structures. The Hague: Mouton.
Chomsky, Noam. 1965. Aspects of the Theory of Syntax. Cambridge, MA: MIT
Press.
Chomsky, Noam. 1970. Remarks on nominalization. In Roderick Jacobs and
Peter Rosenbaum (eds.), Readings in English Transformational Grammar,
184–221. Waltham, MA: Ginn & Co.
Chomsky, Noam. 1986. Knowledge of Language: Its Nature, Origin, and Use. New
York: Praeger.
Diedrichsen, Elke. 2004. The German ‘bekommen-passive’ and RRG. In Brian
Nolan (ed.), Linguistic Theory and Practice: Description, Implementation and
Processing. Proceedings of the 2004 RRG Dublin Conference, 49–72.
Diedrichsen, Elke. 2011. The theoretical importance of constructional sche-
mas in RRG. In Nakamura (ed.), 169–197.
Dixon, R. M. W. 1972. The Dyirbal Language of North Queensland. Cambridge:
Cambridge University Press.
Dixon, R. M. W. (ed.). 1976. Grammatical Categories in Australian Languages.
Canberra: Australian Institute of Aboriginal Studies.
Dixon, R. M. W. 1984. The semantic basis of syntactic properties. BLS 10:
583–595.
Dixon, R. M. W. 1991. A New Approach to English Grammar, on Semantic Principles.
Oxford: Oxford University Press.
Dowty, David. 1979. Word Meaning and Montague Grammar. Dordrecht: Reidel.
Dryer, Matthew. 1986. Primary objects, secondary objects, and antidative.
Language 62: 808–845.
Everett, Caleb. 2006. Patterns in Karitiâna: Articulation, Perception, and Grammar.
PhD dissertation, Rice University, TX [available on Information Structure in
Amazonian languages website].
Everett, Caleb. 2008. Constituent Focus in Karitiâna. Unpublished ms. [available
on Information Structure in Amazonian languages website].
Everett, Daniel. 2002. Notes on an RRG Theory of Morphology. Unpublished ms.
Everett, Daniel. 2005. Cultural constraints on grammar and cognition in
Piraha:~ Another look at the design features of human language. Current
Foley, William A. and Robert D. Van Valin, Jr. 1984. Functional Syntax and
Universal Grammar. Cambridge: Cambridge University Press.
Franklin, Karl. 1971. A Grammar of Kewa, New Guinea. (Pacific Linguistics C-16).
Canberra: Australian National University.
Goldberg, Adele. 2009. Constructions at work. Cognitive Linguistics 20:
201–224.
González Vergara, Carlos. 2006. Las construcciones no reflexivas con ‘se’: Una
propuesta desde la Gramática del Papel y la Referencia. PhD dissertation, Uni-
versidad Complutense de Madrid [available on the RRG website, https://
rrg.caset.buffalo.edu/].
González Vergara, Carlos. 2009. One rule to rule them all: Logical structures
for Spanish non-reflexive se sentences. In Guerrero Valenzuela, Ibáñez and
Belloro (eds.), 361–379.
González Vergara, Carlos. 2011. Se-incompatible predicates in Spanish:
A RRG explanation. In Nakamura (ed.), 134–142.
Guerrero Valenzuela, Lilián, Sergio Ibáñez and Valeria Belloro (eds.). 2009.
Studies in Role and Reference Grammar. México: UNAM.
Guerrero Valenzuela, Lilián and Robert D. Van Valin, Jr. 2004. Yaqui and the
analysis of primary object languages. International Journal of American Lin-
guistics 70: 290–319.
Hale, Kenneth. 1976. The adjoined relative clause in Australia. In Dixon (ed.),
78–105.
Hasegawa, Yoko. 1996. A Study of Japanese Clause Linkage: The Connective TE in
Japanese (Studies in Japanese Linguistics 5). Stanford, CA: CSLI Publications.
Haviland, John. 1979. How to talk to your brother-in-law in Guugu Yimid-
hirr. In T. Shopen (ed.), Languages and Their Speakers, 160–239. Cambridge,
MA: Winthrop.
Himmelmann, Nikolaus. 2008. Lexical categories and voice in Tagalog. In
Paul Kroeger (ed.), Voice and Grammatical Relations in Austronesian Languages,
247–293. Stanford, CA: CSLI Publications.
Holisky, Dee Ann. 1987. The case of the intransitive subject in Tsova-Tush
(Batsbi). Lingua 71: 103–132.
Hudson, Richard. 1992. So-called ‘double objects’ and grammatical relations.
Language 68: 251–276.
Humboldt, Wilhelm von. 1836. Über die Verschiedenheit des menschlichen Sprach-
baues und ihren Einfluss aft die geistige Entwicklung des Menschengeschlects.
Berlin: Dümmler.
Hyman, Larry and Karl Zimmer. 1976. Embedded topic in French. In Charles
Li (ed.), 189–211.
Ikegami, Yoshihiko. 1985. ‘Activity’ – ‘accomplishment’ – ‘achievement’ –
A language that can’t say ‘I burned it, but it didn’t burn’ and one that can.
In A. Makkai and A. Melby (eds.), Linguistics and Philosophy: Essays in honor of
Rulon S. Wells, 265–304. Amsterdam: John Benjamins.
Imai, Shingo. 1998. Logical Structures and Case Marking in Japanese. MA thesis,
University at Buffalo (SUNY) [available on RRG website, https://rrg.caset.
buffalo.edu/].
Ruhnau, Arne. 2011. Interpretation of the Topological Field Model of the German Clause
in Terms of Role and Reference Grammar. MA thesis, Heinrich Heine University
Düsseldorf [available on RRG website, https://rrg.caset.buffalo.edu/].
Sag, Ivan A. 2012. Sign-Based Construction Grammar: An informal synopsis.
In Hans C. Boas and Ivan A. Sag (eds.), Sign-Based Construction Grammar,
69–202. Stanford, CA: CSLI Publications.
Schachter, Paul. 1976. The subject in Philippine languages: Actor, topic,
actor-topic, or none of the above. In Li (ed.), 491–518.
Schachter, Paul and Fe Otanes. 1972. Tagalog Reference Grammar. Berkeley:
University of California Press.
Schwartz, Linda. 1993. On the syntactic and semantic alignment of attribu-
tive and identificational constructions. In Van Valin (ed.), 433–463.
Shimojo, Mitsuaki. 1995. Focus Structure and Morphosyntax in Japanese: wa and
ga, and Word Order Flexibility. PhD dissertation, University at Buffalo (SUNY)
[available on the RRG website, https://rrg.caset.buffalo.edu/].
Shimojo, Mitsuaki. 2011. The left periphery and focus structure in Japan-
ese. In Wataru Nakamura (ed.), New Perspectives in Role and Reference Gram-
mar, 266–293. Newcastle upon Tyne: Cambridge Scholars Publishing.
Silverstein, Michael. 1976. Hierarchy of features and ergativity. In Dixon
(ed.), 112–171.
Smith, Carlotta. 1997. The Parameter of Aspect (2nd ed.). Dordrecht: Reidel.
Talmy, Leonard. 2000. Toward a Cognitive Semantics. Cambridge, MA: MIT Press.
Toratani, Kiyoko. 2002. The Morphosyntax and Logical Structures of Compound
Verbs in Japanese. PhD dissertation, University at Buffalo (SUNY) [available
on RRG website, https://rrg.caset.buffalo.edu/].
Ullrich, Jan. 2011. New Lakota Dictionary (2nd ed.). Bloomington, IN: Lakota
Language Consortium.
Ullrich, Jan. 2018. Modification, Secondary Predication and Multi-Verb Construc-
tions in Lakota. PhD dissertation, Heinrich Heine University Düsseldorf
[available on RRG website, https://rrg.caset.buffalo.edu/].
Ullrich, Jan and Ben Black Bear, Jr. 2018. Lakota Grammar Handbook (2nd ed.).
Bloomington, IN: Lakota Language Consortium.
Van Hooste, Koen. 2018. Instruments and Related Concepts at the Syntax–Semantics
Interface. PhD dissertation, Heinrich Heine University Düsseldorf [available
on RRG website, https://rrg.caset.buffalo.edu/].
Van Valin, Robert D., Jr. 1977. Aspects of Lakhota Syntax. PhD dissertation,
University of California, Berkeley.
Van Valin, Robert D., Jr. 1981. Grammatical relations in ergative languages.
Studies in Language 5: 361–394.
Van Valin, Robert D., Jr. 1985. Case marking and the structure of the Lakhota
clause. In Johanna Nichols and Anthony Woodbury (eds.), Grammar Inside
and Outside the Clause, 363–413. Cambridge: Cambridge University Press.
Van Valin, Robert D., Jr. 1990a. Semantic parameters of split intransitivity.
Language 66: 221–260.
Van Valin, Robert D., Jr. 1990b. Functionalism, anaphora and syntax. (Review
article on S. Kuno, Functional Syntax.) Studies in Language 14: 169–219.
Van Valin, Robert D., Jr. 1991. Another look at Icelandic case marking and
grammatical relations. Natural Language and Linguistic Theory 9: 145–194.
Van Valin, Robert D., Jr. (ed.). 1993. Advances in Role and Reference Grammar.
Amsterdam: John Benjamins.
Van Valin, Robert D., Jr. 1993. A synopsis of Role and Reference Grammar. In
Van Valin (ed.), 1–164.
Van Valin, Robert D., Jr. 1998. The acquisition of wh-questions and the
mechanisms of language acquisition. In Michael Tomasello (ed.), The New
Psychology of Language: Cognitive and Functional Approaches to Language Struc-
ture, 221–49. Mahwah, NJ: LEA [available on RRG website, https://rrg.caset.
buffalo.edu/].
Van Valin, Robert D., Jr. 1999. Generalized semantic roles and the syntax–
semantics interface. In F. Corblin, C. Dobrovie-Sorin and J.-M. Marandin,
(eds.), Empirical Issues in Formal Syntax and Semantics 2, 373–389. The Hague:
Thesus [available on RRG website, https://rrg.caset.buffalo.edu/].
Van Valin, Robert D., Jr. 2001. The acquisition of complex sentences: A case
study in the role of theory in the study of language development. Chicago
Linguistic Society Proceedings 36(2): 511–531.
Van Valin, Robert D., Jr. 2002. The development of subject–auxiliary inver-
sion in English wh-questions: An alternative analysis. Journal of Child Lan-
guage 29: 161–175.
Van Valin, Robert D., Jr. 2003. Some remarks on the bekommen-‘passive’ in
German. Invited paper presented at the 9th German Linguistics Association
Conference, University at Buffalo (SUNY).
Van Valin, Robert D., Jr. 2004. Semantic macroroles in Role and Reference
Grammar. In Rolf Kailuweit and Martin Hummel (eds.), Semantische Rollen,
62–82. Tübingen: Narr [available on RRG website, https://rrg.caset.buffalo.
edu/].
Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cam-
bridge: Cambridge University Press.
Van Valin, Robert D., Jr. 2006. Semantic macroroles and language processing.
In Ina Bornkessel, Matthias Schlesewsky, Bernard Comrie and Angela D.
Friederici (eds.), Semantic Role Universals and Argument Linking: Theoretical,
Typological, and Psycholinguistic Perspectives, 263–301. Berlin: Mouton de
Gruyter.
Van Valin, Robert D., Jr. (ed.). 2008. Investigations of the Syntax–Semantics–
Pragmatics Interface. Amsterdam: John Benjamins.
Van Valin, Robert D., Jr. 2008. RPs and the nature of lexical and syntactic
categories in RRG. In Van Valin (ed.), 161–178.
Van Valin, Robert D., Jr. 2009. Privileged syntactic arguments, pivots, and
controllers. In Guerrero Valenzuela, Ibáñez and Belloro (eds.), 45–68.
Van Valin, Robert D., Jr. 2012a. Lexical representation, co-composition, and
linking syntax and semantics. In James Pustejovsky, Pierrette Bouillon,
Hitoshi Isahara, Kyoko Kanzaki and Chungmin Lee (eds.), Advances in
Generative Lexicon Theory, 67–107. Dordrecht: Springer.
Van Valin, Robert D., Jr. 2012b. Some issues in the linking between syntax
and semantics in relative clauses. In Bernard Comrie and Zarina Estrada
Fernández (eds.), A Typological Overview of Relative Clauses in Languages of the
Americas, 47–64. Amsterdam: John Benjamins.
Van Valin, Robert D., Jr. 2013. Head-marking languages and linguistic theory.
In Balthasar Bickel, Lenore A. Grenoble, David A. Peterson and Alan
Timberlake (eds.), Language Typology and Historical Contingency. In honor of
Johanna Nichols, 91–124. Amsterdam: John Benjamins.
Van Valin, Robert D., Jr. 2016. An overview of information structure in three
Amazonian languages. In M. M. Jocelyne Fernandez-Vest and Robert D. Van
Valin, Jr. (eds.), Information Structuring of Spoken Language from a Cross-
Linguistic Perspective, 77–92. Berlin: Mouton de Gruyter.
Van Valin, Robert D., Jr. 2018a. Some issues regarding (active) accomplish-
ments. In Rolf Kailuweit, Lisa Künkel and Eva Staudinger (eds.), Applying
and Expanding Role and Reference Grammar, 71–94. Freiburg: FRIAS.
Van Valin, Robert D., Jr. 2018b. Dative case and oblique subjects. In Jóhanna
Barðdal, Steven Carey, Thórhallur Eythórson and Na’ama Pat-El (eds.), Non-
Canonically Case-Marked Subjects: The Reykjavík-Eyjafjallajökull Papers, 115–131.
Amsterdam: John Benjamins.
Van Valin, Robert D., Jr. (ed.). 2021. Challenges at the Syntax–Semantics–
Pragmatics Interface: A Role and Reference Grammar Perspective. Newcastle upon
Tyne: Cambridge Scholars Publishing.
Van Valin, Robert D., Jr. 2021. Cosubordination. In Van Valin (ed.), 241–283.
Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning
and Function. Cambridge: Cambridge University Press.
Van Valin, Robert D., Jr. and David P. Wilkins. 1993. Predicting syntactic
structure from semantic representations: remember in English and
Mparntwe Arrernte. In Van Valin (ed.), 499–534.
Van Valin, Robert D., Jr. and David P. Wilkins. 1996. The case for ‘effector’:
Case roles, agents and agency revisited. In M. Shibatani and S. Thompson
(eds.), Grammatical Constructions, 289–322. Oxford: Oxford University Press.
Vendler, Zeno. 1967. Linguistics in Philosophy. Ithaca, NY: Cornell University
Press.
Walton, Charles. 1986. Sama Verbal Semantics: Classification, Derivation and
Inflection. Manila: Linguistic Society of the Philippines.
Watters, James K. 1993. An investigation of Turkish clause linkage. In Van
Valin, (ed.), 535–560.
Winther-Nielsen, Nicolai. 1995. A Functional Discourse Grammar of Joshua:
A Computer-Assisted Rhetorical Structure Analysis. Stockholm: Almqvist &
Wiksell.
Winther-Nielsen, Nicolai. 2021. Why Eve shouldn’t eat the snake: An intelli-
gent answer from corpus-driven information structure and reference
tracking in Biblical Hebrew. In Van Valin (ed.), 285–307.
Yang, Byong-seon. 1994. Morphosyntactic Phenomena of Korean in Role and Refer-
ence Grammar: Psych-Verb Constructions, Inflectional Verb Morphemes, Complex
Notes
1 In the tree diagrams modifiers are related to the units modified via
arrows. Kallmeyer and Osswald (Chapter 20) propose an alternative for-
malization of the layered structure of the clause which does not use
arrows to represent adjunct modifiers. I will continue to use the trad-
itional RRG representations.
2 Bohnemeyer and Van Valin (2017) show that the core is the unit of the LSC
which expresses a single event; it has the macro-event property, which the
nucleus and the clause lack. See Section 1.3.4.4.
3 For explications of these notions, see Bentley (Chapter 11).
4 The auxiliary did is not attached to anything in the tree in Figure 1.4. Its
status will be addressed in Section 1.2.3. With respect to the PrCS and
resumptive elements, the claim that PrCS XPs never co-occur with a
resumptive element holds true for English and many other languages.
However, in head-marking languages (see Section 1.2.7.2) which have a
PrCS, e.g. Tzotzil (Aissen 1987), an argument in the PrCS is cross-
referenced on the verb, which could be viewed as a type of resumptive
element; what would be impossible would be for there to be an inde-
pendent pronoun in the clause referring to the PrCS RP. Spanish pre-
sents an interesting case as well. Dative arguments are obligatorily
realized by a dative clitic on the verb and optionally also by a dative-
marked noun or pronoun (see (58) in Section 1.2.7.2); in wh-questions in
which the wh-expression is dative, such as (58d), it occurs in the PrCS
and is still cross-referenced by the dative clitic in the core. Hence the
issue of resumption and the PrCS is more complex than it appears to be
in English.
5 It’s been claimed by a number of linguists, most notably Kayne (1994) in
his theory of the antisymmetry of syntax, that all displacement is to the
beginning of the sentence. These Dhivehi examples are a counterexample
to this claim.
6 I am grateful to colleagues from Iran (Dr Mozhgan Neisani, Dr Zahra
Ghane’) for bringing this problem to my attention. See Winther-Nielsen
(2021) for the same problem regarding the analysis of Biblical Hebrew. The
post-detached position (PoDP) involves an intonation break and resump-
tive cataphoric pronoun for arguments, e.g. What did shei eat after lunch,
Maryi?
7 ‘DetP’ will be used for ‘determiner phrase’ rather than the usual ‘DP’,
because ‘DP’ has already been used to stand for ‘detached position’ in
PrDP and PoDP.
8 See Section 1.2.7.2 for an explanation as to why the independent RP is
not core-internal.
9 In Van Valin (2005) ‘RP’ stands for ‘referential phrase’, but this term is
problematic for various reasons (see Peterson, Chapter 2 of this volume).
In Van Valin (2008) it was replaced by ‘reference phrase’ and defined as a
potentially referring expression, thereby taking into account the fact that
RPs can be non-referential in certain contexts.
10 RP, MP and nucleus are universal categories, while PP is not, since there
are languages which lack adpositions altogether, e.g. Dyirbal.
11 See Peterson (this volume, Chapter 2) for an in-depth discussion of
this topic.
12 Peterson (Chapter 2) argues that Kharia, a South Munda language (India),
makes no categorial distinctions among content morphemes at all.
13 The existence of attributive modifiers in Lakhota has been controversial
since Boas and Deloria (1941). See Ullrich (2018) for a detailed analysis of
modification in Lakhota.
14 Bohnemeyer (2019) presents an alternative conception of operators and
their relationship to the LSC based on a different theory of tense and
aspect from the one assumed traditionally in RRG.
15 The formalization of RRG proposed in Kallmeyer and Osswald (Chap-
ter 20) does away with the operator projection and represents both
constituents and operators in the same tree structure. I will continue
to use the traditional representations.
16 Lakhota lacks tense marking, so this sentence can be interpreted as
present or past, depending on context.
17 This is not the complete array; see Ullrich (2011: 820–822) for a full list.
18 See Peterson (Chapter 2) for more discussion and exemplification of
this claim.
19 See O’Connor (2008) for the outlines of a prosodic projection for the LSC
in which the intonational properties of different sentence types and
constructions can be represented.
20 Additional examples of attachment to more than one projection will be
given later in this section and in Section 3.
21 See Toratani (Chapter 9) for an in-depth discussion of adverbs and Ibáñez
Cerda (Chapter 10) for a detailed discussion of adpositional phrases.
22 If the adjunct PP were treated as a parenthetical, then Max drank, after the
argument with Mary, an entire bottle of wine would be acceptable, but it is
highly marked and not the same structure as (30b).
23 For detailed discussion of the ordering of constraints on adverbs, see
Toratani (Chapter 9).
24 Unlike Japanese, German is not consistently left-branching; it has many
more prepositions than postpositions, and in complex sentences
Abbreviations
2.1 Introduction
The present chapter deals with the status of lexical and grammatical cat-
egories in RRG, including parts-of-speech systems and grammatical
categories such as tense, aspect, mood (TAM) and negation, which are
referred to in RRG as ‘operators’.* With respect to lexical categories, RRG
differs from many other theories of language which assume a small, closed
set of universal parts of speech such as ‘noun’, ‘verb’ ‘adjective’, etc., and
which consider these to be the heads of corresponding syntactic categories,
namely NP, VP and AdjP. By contrast, RRG does not assume that such
categories are universal, in line with an increasing amount of typologically
oriented research questioning the universality of these categories. Instead,
RRG assumes functionally motivated, non-endocentric syntactic categories,
such as nucleus or NUC (containing the predicate), referential phrases or
RPs, and modifying phrases or MPs. Although these syntactic slots are
typically realized by verbs, nouns and adjectives/adverbs, respectively, this
is not required by the theory. Many languages, including English, allow for
non-verbal predicates, non-adjectival modifying phrases, etc., while others
show little or no evidence for lexical categories such as ‘noun’ and ‘verb’ or
allow for clausal referents, modifiers and predicates. And even those lan-
guages which do possess traditional parts of speech often show so-called
‘category squishes’, in which assignment to a particular category is at best
an arbitrary decision. Hence, categories such as NUC, RP and MP are not
universally linked to particular lexical categories per se, although individ-
ual languages will generally have language-specific restrictions as to which
elements may occupy these slots.
The present chapter also provides a description of the grammatical cat-
egories referred to in RRG as operators, that is, those categories which
‘ground’ a clause, core or nucleus and which are closely linked to finiteness,
a topic which has been receiving increasing attention in typological
research in recent years. These categories include tense, aspect, deontic
mood (‘modality’ in RRG terms), epistemic mood (‘status’), evidentials, etc.
These units all have scope over a particular level of the sentence structure
(nucleus, core, clause) and play an important role in the determination of
nexus relations in complex sentences within the theory. This also includes
categories which are primarily concerned with questions of reference, such
as definiteness and deixis, which ground the RP.
* I would like to thank Utz Maas for his comments on an earlier version of the present study. Needless to say, any remaining
errors and misconceptions are my own.
as noun, verb and adjective (to name just three) as well as the correspond-
ing endocentric syntactic categories, NP, AdjP and VP, even though
these categories – or at least the lexical categories – are generally
considered universal.
Although there are considerable differences of opinion with respect to the
‘correct’ approach to this topic, most researchers would probably agree with
an approach which views lexical categories as language-specific categories
based on feature bundles, in which ‘prototypical’ members of a particular
class possess all of the features potentially associated with the particular
class.2 For example, prototypical nouns in English would denote persons,
places or things, mark for the plural (usually with -s), be compatible with
definite and indefinite articles, etc., while less prototypical nouns, for
example, may not be compatible with plural marking or the definite article.
Problematic for this approach, however, is determining which features are
more important than others, if indeed a hierarchy can be established at all,
as well as how many features are ‘enough’ for an item to be assigned to a
particular category. We will return to this topic in the following pages.
As Himmelmann (2008: 260) notes, there is little doubt that ‘the inventory
of function words is highly language-specific’. Rather, ‘[w]henever there is a
controversy regarding the number and kind of syntactic categories in a
given language, it pertains to the linguistic classification of content words’.
Himmelmann assumes two grammatical levels for the discussion of parts of
speech: (i) ‘the level of terminal syntactic categories where lexical items are
categorised according to their phrase-structural properties’, and (ii) ‘the
level of lexical categories proper where lexical items are categorised
according to those grammatical features which are not directly relevant
for phrase structure’ (Himmelmann, 2008: 263). To illustrate these two
levels,3 Himmelmann proposes a hypothetical example in a language L, in
which the most basic phrasal categories always consist of an overt function
word X and an overt content word Y, with virtually all content words being
compatible with all function words. This can be schematically portrayed as
in (1). Here the status of the function word X distinguishes the phrase from
other phrase types, hence ‘XP’.
These two levels, that is, lexical and syntactic, need not correspond to one
another directly, although this will often be the case, as, for example, with
Ns as heads of NPs, Adjs as heads of AdjPs and Vs as heads of VPs in many
languages. Himmelmann (2008: 264) notes five logical possibilities with
respect to these two levels and their correspondence to one another, given
in Table 2.1. With respect to category I, both levels possess distinct categories
but these do not directly correlate in Ia (‘6¼’), while they do in Ib (‘¼’).
Returning to our example above, if, following Himmelmann, we imagine
that L has a further restriction such that not all content words Y behave
similarly, as some Ys take plural suffixes, others take plural prefixes, while
Ia distinct 6¼ distinct
Ib distinct ¼ distinct
II indistinct distinct
III distinct indistinct
IV indistinct indistinct
the members of a third class do not allow pluralization, then we would have
case Ia, with distinct lexical categories based on the compatibility of Y with
plural markers and also distinct terminal syntactic categories, but with no
direct correspondence between the two levels, as the compatibility of plural
affixes does not correlate with the compatibility of a particular Y with a
particular X.
To these two descriptive levels we can add a third level, that of function,
here primarily reference, predication and attribution. This can involve a
direct correlation between terminal syntactic cateogies and functional cat-
egories, that is, XP1 ¼ reference, XP2 ¼ predication, etc., or there can be no
direct correlation, so that XP1, XP2, etc. can both be used in reference,
predication and modification. At one extreme we would then have endo-
centric categories specialized for certain sentence-level functions, for
example N ¼ NP ¼ reference, V ¼ VP ¼ predication, etc., and at the other
extreme – at least putatively – no correlations whatsoever, although it is
questionable whether such a language is possible.
Such mismatches are not only a theoretical possibility; many research-
ers have claimed that the languages they are describing either do not in
fact possess a certain category such as ‘N’ or ‘NP’, or that the respective
language can at least best be described without recourse to such categor-
ies. Even in languages such as English, which possess relatively clearly
defined lexical classes such as noun, verb, and adjective, it has long been
known that there is no one-to-one relationship between part of speech
and function. Consider the examples in (2a–g), from Van Valin
(2008: 165).
(2) a. Chris will [nuc [pred see]] the movie. PRED in NUC ¼ V
b. Chris [nuc is [pred a very good detective]]. PRED in NUC ¼ NP
c. Pat [nuc is [pred exceedingly tall]]. PRED in NUC ¼ ADJP
d. Pat [nuc is [pred in the house]]. PRED in NUC ¼ PP
e. Chris [nuc [pred wiped]] PRED in NUC1 ¼ V
the table [nuc [pred squeaky clean]]. PRED in NUC2 ¼ ADJP
f. Pat [nuc [pred pushed]] PRED in NUC1 ¼ V
the table [nuc [pred out the door]]. PRED in NUC2 ¼ PP
g. Chris [nuc was [pred elected]] PRED in NUC1 ¼ V
[nuc [pred president of the club]] PRED in NUC2 ¼ NP
SENTENCE SENTENCE
CLAUSE CLAUSE
CORE CORE
RP NUC RP RP NUC
V NP
SENTENCE
CLAUSE
CORE
RP NUC
AUX PRED
PP
As the examples in (2) show, the predicate of a nucleus in English need not
be verbal. While the predicate is verbal in (2a), in (2b) it is an NP, in (2c) an
AdjP and in (2d) a PP. Figure 2.1 presents the constituent projections for (2a)
(from this author), and (2b) and (2d) (adapted from Van Valin, 2008: 166).4
In English, as in many other languages, there is a general requirement
that clauses in which the predicate is not verbal must contain a kind of
‘auxiliary’ such as the copula is in order for the clause to be grammatical. As
a very good detective in (2b), exceedingly tall in (2c) and in the house in (2d) are not
verbs, the copula is is required; however, it is important to stress two things
here. First, that while is is a verb, it is not the predicate; this function is filled
by the NP in (2b), by the AdjP in (2c), and by the PP in (2d). Second, this is a
language-specific criterion which is not found in all languages or which is
only found with certain interpretations. For example, one quite common
situation is found in Russian, where a clause whose predicate is non-verbal
does not require a copula when there is a present or atemporal interpret-
ation, as in (3).
(3) Russian
eto stol
this table
‘This [is a] table.’
As Van Valin (2008: 166) notes, non-verbal predicates do not pose a prob-
lem in RRG, since the clause is not an endocentric category: ‘Indeed, there is
no head at all. The nucleus cannot be considered the “head” of the core or
the clause, because it is not a lexical category, on the one hand, and is often
phrasal, on the other. The notion of “head” is of no relevance to the layered
structure of the clause.’ Instead, all universal categories assumed in RRG are
semantically motivated, so that in RRG a clause minimally consists of a
core – which minimally consists of a nucleus which contains the predicate –
and possibly one or more arguments.
Instead of positing universal lexical categories which are also the heads of
corresponding phrasal categories – cross-linguistically an even more prob-
lematic assumption – mainstream RRG has assumed since Van Valin (2008)
that the core contains a predicate, the functional/semantic base of the
nucleus, and possibly one or more referential phrases, referred to as RPs.
Just as the predicate need not necessarily be verbal, neither does an RP have
to be nominal. In addition, RRG assumes that these units may be modified
by modifying phrases or MPs, which can (but need not) be specialized so as
to modify only an RP, as with prototypical adjectives, or only predicative
elements, as with prototypical adverbs, although again this is not necessary,
as there are many languages with either no or only very few adjectives and/
or adverbs, where these functions are realized by other categories, such as
nouns or verbs.
Not all NPs are RPs and not all adjectives or adverbs are MPs. For
example, expletive or ‘dummy’ subjects like those in (5a) are not referen-
tial and hence not RPs: these are pronouns and therefore NPs, but they are
not referential and hence not RPs. Similarly, while a very good detective in
example (2b) is an NP, repeated here as (5b), it is not referential and
therefore also not an RP.
The same holds for the relationship between adjectives / adjective phrases
and the MP. For example, while the AdjP very good in (5b) is a modifier in the
NP a very good detective, and hence an MP, the AdjP very good in (6) is not a
modifier, and therefore not an MP.
In languages which do not have the usual parts of speech such as noun,
adjective or verb, these units will of course belong to different, language-
specific categories. For example, as we will see in Section 2.2.3.1 on Kharia,
translation equivalents of nouns, adjectives and verbal stems all belong to
the open class of contentive lexemes of that language, whereas NPs and
finite verbs in English correspond most closely to the Kharia-specific units
‘Case-syntagma’ and ‘TAM/Person-syntagma’, respectively. Again, these struc-
tural units must be carefully distinguished from the functionally motivated
units RP, MP and predicate.
As mentioned above, modifying phrases or MPs in many languages have
as their default value either adjectives or adverbs, depending on what is
being modified, that is, a noun (the tall tree) or a verb (the mouse ran quickly
into the closet). These appear in the peripheries of RPs and clauses (Van Valin
2008: 172).5 As Van Valin notes, MPs also have a layered structure, for two
reasons. First, many languages allow modifier phrases with an adjectival
nucleus to take a core argument, such as the German der auf seinen Sohn
stolze Vater (lit. ‘the of his son proud father’). Secondly, the modifiers can
themselves be modified, ‘which means that they must have a periphery to
house the modifying MP, e.g. the very quickly extinguished fire’ (Van Valin
2008: 172).
In addition to adjectives and adverbs, many languages, including English,
also allow other, much more complex categories to serve as modifiers.
Consider the following examples, from the same source.
The status of this unit in (8) is markedly different from that of the under-
lined units in (9), from the same source, referred to as non-predicative PPs as
RP
CORER
PERIPHERY NUCR
MP N
COREM
NUCM
SENTENCE
CLAUSE PoDP
RP NUC MP
ADV MP MP N
these units do not predicate but are licensed by the verb and therefore do
not license an object.
a. PP b. PP
COREP
P RP NUCP RP
NUCR P NUCR
N N
Although proved in example (10a) is clearly verbal and proof in the (10f)
examples is clearly nominal, the status of the forms ending in -ing is much
less clear. While some instances of proving, such as those in (10c) and (10d),
take a direct object, suggesting that this form is a verb, the forms in (10e) and
(10e′′) require an ‘object’ marked by the preposition of, suggesting that this is
a noun. Furthermore, the ‘subject’ NP in (10c) appears as a bare NP, while in
(10d) – as in the more ‘nominal’ form in (10e) or the clearly nominal form
proof in the (10f) examples – it appears in the genitive. These and other factors
such as adjectival modification, optional complements, tense, aspect and
voice are summarized in tabular form in Van Valin (2008: 175).6
As we see in (10), even in languages such as English, which otherwise seem
to have quite clear lexical classes, it is often not feasible to assume that each
and every word form unambiguously belongs to a particular lexical
category. RRG thus assumes that it is the individual features which are
relevant and not the labels themselves:
Lexicon
(11) Kharia ((11a) from Peterson (2011a: 76); (11b) adapted from Malhotra (1982: 136))
a. lebu ɖel¼ki.
man come¼mid.pst
‘The/a man came.’
b. bhagwan lebu¼ki ro ɖel¼ki.
God man¼mid.pst and come¼mid.pst
‘God became man [¼ Jesus] and came [to earth].’
am¼ga am¼na¼m.’
2sg¼foc 2sg¼mid.irr¼2sg
‘“In the play I will be him and you will be me.” “No. I can’t be you. The director already
made me him. You will be you.”’
(16) Properties: rusuŋ ‘red (one); red (attributive); become red’, maha ‘big (one);
big; grow, become big’
(18) Status and Role: ayo ‘mother; become a mother (middle voice), accept
someone as a mother (active voice)’
(19) Deictics and proforms: iɖaʔ ‘yesterday; become yesterday (middle), turn (e.g.
today) into yesterday (active, e.g. with God as subject)’
(20) Physical objects and animate entities: kaɖoŋ ‘fish; become a fish (middle),
turn into a fish (active)’
On the basis of data such as these, Peterson (2013) assumes two structur-
ally defined categories, the Case-syntagma and the TAM/Person-syntagma,
both of which have the same potential structure for the semantic base and
which differ only with respect to their functional marking, namely case or
TAM/voice and person/number/honorific marking, motivated by similar
structures assumed for Tongan by Broschart (1997). The structure of the
semantic base is given in (27). None of these elements is obligatory, as long
as some non-enclitic unit is present. The Kleene star (*) denotes that poten-
tially any number of content morphemes (lex) is possible, including zero.
(28) X¼case
Case in (28) refers to postpositions such as buŋ ‘with; ins’, seŋ ‘before’, etc.,
as well as the two cases given in (29), so that any postposition, ¼te, or the
lack of any overt marking signals that the unit is a Case-syntagma.
(29) Case: Direct (zero marking)7 the case of subjects and indefinite direct objects;
Oblique (marked by ¼te) marks definite direct objects, ‘indirect
objects’ and adverbials
The genitive is not a case in this sense: whereas the direct and oblique
cases are relevant at the clause level to mark the relation of a constituent to
the predicate (roughly: subject/non-subject), the genitive is only relevant
within the semantic base and serves to integrate one semantic base into a
larger semantic base. The oblique case also cannot appear in a TAM/Person-
syntagma, as (30) shows, whereas the genitive can (31).
CASE-SYNTAGMA
TAM/PERSON-SYNTAGMA
These two syntactic units are defined purely structurally, and both can
appear in attributive, predicative and referential functions, as examples
(39)–(42) from Peterson (2013: 137) show.
(39) Kharia
kuda koloŋ daru
millet bread Tree
‘a millet bread tree’ (name of a children’s story)
(Note: kuda modifies koloŋ and the two together modify daru.)11
However, if the semantic part of this unit is re-ordered as in (45), with two
constituents instead of just the one in (43), the unit which is not marked by
a postposition may serve as the semantic base of a TAM/Person-syntagma.
This is a common predicate type in Kharia (Peterson 2011a: 220).
At issue here is that the unit ending in the postposition may not be directly
followed by a TAM marker, whereas the unit which does not end in a postpos-
ition/case marker can. Furthermore, if a ‘light verb’ such as hoy ‘become’
(borrowed from Kharia’s Indo-Aryan neighbour Sadri) is inserted between buŋ
‘ins’ and goʔɖ¼ki in (44), this results in a grammatical predicate, consisting of
a Case-syntagma and a marker of qualitative predication ( copula).13
RRG can thus easily account for analyses of individual languages which do
not have supposedly universal categories such as ‘noun’, ‘adjective’ and/or
‘verb’, without needing any empty categories or hidden verbs.
Finally, there is a rather marginal type of TAM/Person-syntagma in Kharia,
quotatives, in which the semantic base consists of an entire sentence, as in
example (48).
SENTENCE
PERIPHERY CLAUSE
RP PERIPHERY CORE
CASE- TAM/PERSON-
SYNTAGMA SYNTAGMA
CORE P
CASE-
SYNTAGMA RP NUCP
In sentences in which the verb is not the first sentence constituent, for
example with constituent questions, this sentence-initial unit is followed
by a different class of clitics which mark tense and which agree with the
gender of the item in sentence-initial position, regardless of its function in
the sentence, as in (51).
If the ISC predicate is considered the predicate, just like verbs in verb-initial
sentences, the following generalizations may be made for constituent order
in Wari’ (Everett 2008: 392):
SENTENCE
CLAUSE
CORE
NUC RP RP
PRED
CLAUSE
PreCore CORE
NUC PRO RP
NP PRED
Sentence
Predicate A1, A2, A3, … Speech act type Time Questions of reference …
(command / of referential
request, declarative, state / identifiable
question, …) event …
Figure 2.8 The sentence base and sentence modality
In RRG, this information is divided into two broad categories: the first group
we will refer to as ‘proposition-grounding’, as these categories ground the
event or state with respect to time, modality, speech act, etc. The second
group we will refer to as ‘referent-grounding’, as it refers to
referential identifiability.
Table 2.2 Operators in the layered structure of the clause, from Van Valin (2005: 9)
Nuclear operators:
Aspect
Negation
Directionals (only those modifying the orientation of an action or event without reference to the
participants)
Core operators:
Directionals (only those expressing the orientation or motion of one participant with reference to
another participant or to the speaker)
Event quantification
Modality (root modals, e.g. ability, permission, obligation)
Internal (narrow scope) negation
Clause operators:
Status (epistemic modals, external negation)
Tense
Evidentials
Illocutionary force (IF)
SENTENCE SENTENCE
CLAUSE CLAUSE
CORE CORE
RP NUC RP NUC
V PP
V PP
IF CLAUSE IF CLAUSE
SENTENCE SENTENCE
Figure 2.9 Constituent and operator projections for two English sentences (adapted from
Van Valin 2005: 14)19
SENTENCE
CLAUSE
PERIPHERY CORE
RP RP NUC
PRED
PP V
CORE
CLAUSE TNS
CLAUSE IF
SENTENCE
Figure 2.10 Constituent and operator projections in a Japanese sentence (adapted from
Van Valin 2005: 14)
b. Íra-pa-saa-ru.
cook-prf-up-1sg.pst
‘I burned it upward (as a hill).’ (V-ASP-DIR-TNS)
b. Gel-emi-yebil-ir-im.
come-able.neg-psbl-aor-1sg
‘I may be unable to come.’ (V-MOD-STA-TNS)
Type Scheme
I Eþ P No (verbal) morphology
Ia Eþ P (Modifying þ Modified) No (verbal) morphology
II Eþ P-finite Morphologically finite predicate
IIa Eþ P (Modifying þ Modified-finite) Modified element (¼ ‘lexical verb’) is finite,
modifying (or ‘situating’) element is
invariable (e.g. ‘particle’)
IIb E þ P (Modifying-finite þ Modified) Modifying element is finite, modified
element is non-finite (¼ auxiliary-
participle type)
IIc E þ P (Modifying-finite þ Modified-finite) Both elements are finite
III (E þ P)-finite Movable affixes
(Maas 2004: 379, modified and reprinted with kind permission by de Gruyter)
E refers to the complement of a predicate, which may be either an argument or an adjunct;
P denotes the predicate; þ refers to a loose concatenation of words and - to a narrow
morpheme concatenation.
As noted above, RRG does not posit a privileged position for such marking,
and markers of these categories can appear almost anywhere in the sentence.
This allows us to account for such cross-linguistic variability in a straightfor-
ward way, without having to resort to movement or other processes to derive
the actual word order in a particular language. In fact, the only claim that
RRG makes is that the ordering of these morphemes with respect to the
predicate – when they appear on the same side of this unit – corresponds to
the relative scopal properties of the individual operators. And in the case of
the Paez example in (62) this is unproblematic, as the ‘mobile’ markers are
clausal operators which appear in the final position when attached to the
verb in (62a), after aspectual marking and therefore conforming to this
constraint. The position of this marking in (62b) directly preceding the
predicate is also not an exception to this rule, as here the marking of this
clausal operator is not on the same side of the verbal root as the other
operators and hence this constraint does not apply here.
SENTENCE
CLAUSE
CORE
PRED
TAM/PERSON-syntagma
NUC
NUC NUC RP
NUC ASP
CORE
CLAUSE TNS
CLAUSE IF
SENTENCE
Figure 2.11 Constituent and operator projections in Kharia
(63) 〈IF DEC 〈EVID HS 〈TNS PST 〈STA IRR 〈NEG Ø 〈MOD OBLIG 〈EVQ SG 〈DIR Ø 〈ASP PERF
〈LS〉〉〉〉〉〉〉〉〉〉
(64), also from Van Valin (2005: 50), provides an example of the
formal representation of the operators for the English question Has Kim
been crying?.
(64) 〈IF INT 〈TNS PRS 〈ASP PERF PROG 〈do′ (Kim, [cry′ (Kim)])〉〉〉〉
Table 2.4 Operators in the layered structure of the RP (adapted from Van Valin
2005: 24)
NuclearR operators:
Nominal aspect (count–mass distinction, classifiers in classifier languages)
CoreR operators:
Number
Quantification (quantifiers)
Negation
RP operators:
Definiteness
Deixis
RP
CORER
PERIPHERYR NUCR
ADJ N
NUCR
CORER NUM
QNT CORER
DEF RP
Figure 2.12 Layered structure of the RP in English (adapted from Van Valin 2005: 25)
(66) 〈 DEIC PROX 〈 DEF þ 〈 NEG Ø 〈QNT 9 〈 NUM SG 〈 nasp count 〈N〉〉〉〉〉〉〉
The logical structure for a phrase like the scarf would then be represented as
in (67), also from Van Valin (2005: 52).
SENTENCE
CLAUSE
PERIPHERY CORE
N N V N
N N V N
DEIC RP SENTENCE
DEIC RP
Figure 2.13 The structure of the Dyirbal sentence in (65b) (adapted from Van Valin 2005: 29)
(68) 〈IF INT 〈TNS PRES 〈ASP PERF PROG 〈do′ (x, [cry′ (x)] 〉〉〉〉
References
Ambros, Arne A. 1998. Bonġornu, kif int? Einführung in die maltesische Sprache.
Wiesbaden: Reichert Verlag.
Bisang, Walter. 2001. Finite vs. non finite languages. In Martin Haspelmath,
Ekkehard König, Wulf Oesterreicher and Wolfgang Raible (eds.), Language
Typology and Language Universals, Vol. 2, 1400–1413. Berlin: Mouton de
Gruyter.
Broschart, Jürgen. 1997. Why Tongan does it differently: Categorial distinc-
tions in a language without nouns and verbs. Linguistic Typology 1(2):
123–165.
Croft, William. 2000. Parts of speech as language universals and as language-
particular categories. In Petra M. Vogel and Bernard Comrie (eds.),
Approaches to the Typology of Word Classes (Empirical Approaches to Lan-
guage Typology, 23), 65–102. Berlin: Mouton de Gruyter.
Croft, William. 2005. Word classes, parts of speech, and syntactic argumen-
tation. Linguistic Typology 9(3): 431–441.
Evans, Nicholas and Toshiki Osada. 2005. Mundari: The myth of a language
without word classes. Linguistic Typology 9(3): 351–390.
Everett, Daniel L. 2008. Wari’ intentional state constructions. In Van Valin,
Jr. (ed.), 381–410.
Håkansson, Gisela and Jennie Westander. 2013. Communication in Humans and
Other Animals (Advances in Interaction Studies, 4). Amsterdam: John
Benjamins.
Haspelmath, Martin. 2007. Preestablished categories don’t exist: Consequences
for language description and typology. Linguistic Typology 11(1): 119–132.
Haspelmath, Martin. 2011. The indeterminacy of word segmentation and
the nature of morphology and syntax. Folia Linguistica 45(1): 31–80.
Hengeveld, Kees and Jan Rijkhoff. 2005. Mundari as a flexible language.
Linguistic Typology 9(3): 406–431.
Himmelmann, Nikolaus P. 2008. Lexical categories and voice in Tagalog. In
Peter K. Austin and Simon Musgrave (eds.), Voice and Grammatical Relations
in Austronesian Languages (Studies in Constraint-Based Lexicalism),
247–293. Stanford, CA: CSLI Publications.
Hockett, Charles F. 1960. The origin of speech. Scientific American 203: 88–111.
[Reprinted in Wang, William S.-Y. 1982. Human Communication: Language
and Its Psychobiological Bases. Scientific American: 4–12].
Kerkeʈʈa, K. P. 1990. Jujhair ɖā̃ɽ (khaɽiyā nāʈak). Ranchi: Janjātı̄ya Bhāsā
˙
Akādemı̄, Bihār Sarkār.
Klaiman, M. H. 1991. Grammatical Voice (Cambridge Studies in Linguistics,
59). Cambridge: Cambridge University Press.
Luuk, Erkki. 2010. Nouns, verbs and flexibles: Implications for typologies of
word classes. Language Sciences 32(3): 349–365.
Maas, Utz. 2004. ‘Finite’ and ‘nonfinite’ from a typological perspective.
Linguistics 42(2): 359–385.
Malhotra, V. (1982). The Structure of Kharia: a Study of Linguistic Typology and
Language Change. Unpublished PhD dissertation. New Delhi: Jawaharlal
Nehru University.
Nikolaeva, Irina (ed.). 2007. Finiteness. Theoretical and Empirical Foundations.
Oxford: Oxford University Press.
Peterson, John. 2005. There’s a grain of truth in every ‘myth’, or, Why the
discussion of lexical classes in Mundari isn’t quite over yet. Linguistic
Typology 9(3): 391–405.
Peterson, John. 2011a. Kharia. A South Munda language (Brill’s Studies in the
Languages of South and Southwest Asia, 1). Leiden: Brill.
Peterson, John. 2011b. Aspects of Kharia grammar – A Role and Reference
Grammar (RRG) approach. In Rajendra Singh and Ghanshyam Sharma
(eds.), Annual Review of South Asian Languages and Linguistics, 81–124. Berlin:
Mouton de Gruyter.
Peterson, John. 2013. Parts of speech in Kharia: a formal account. In Rijkhoff
and van Lier (eds.), 131–168.
Pustejovsky, James J. 1995. The Generative Lexicon. Cambridge, MA: MIT Press.
Rauh, Gisa. 2010. Syntactic Categories: Their Identification and Description in
Linguistic Theories (Oxford Surveys in Syntax and Morphology, 7). Oxford:
Oxford University Press.
Rijkhoff, Jan and Eva van Lier. 2013. Flexible Word Classes: Typological Studies of
Underspecified Parts of Speech. Oxford: Oxford University Press.
Ross, John Robert. 1972. The category squish: Endstation Hauptwort. In
Papers from the Eighth Regional Meeting. Chicago Linguistic Society. April 14–16,
1972, 316–328.
Roy, Sarat Chandra and Ramesh Chandra Roy. 1937. The Khāriās. Ranchi: Man
˙
in India.
Tomasello, Michael. 2008. Origins of Human Communication (The Jean Nicod
Lectures). Cambridge, MA: MIT Press.
Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cam-
bridge: Cambridge University Press.
Van Valin, Robert D., Jr. 2008. RPs and the nature of lexical and syntac-
tic categories in Role and Reference Grammar. In Van Valin (ed.),
161–178.
Van Valin, Robert D., Jr. (ed.). 2008. Investigations of the Syntax–Semantics–
Pragmatics Interface (Studies in Language Companion Series 105). Amster-
dam: John Benjamins.
Notes
3.1 Introduction
* This research was funded by the Spanish Ministry of Economy and Competitiveness: grants FFI2011–29798-C02–01 and
FFI2014–53788-C3–1-P.
LaPolla 1997) and of active accomplishments (Van Valin, 2005, 2018; cf.
Bentley 2019), these issues still need to be explored in greater depth.
It goes without saying that the most basic function of language is to
convey meaning, and this involves considerably more than syntactic struc-
tures. At some level, an enriched lexical representation system would have
to account for the encyclopedic knowledge that speakers possess, which
enables them to choose the predicate that best represents the meaning they
wish to convey. For example, native speakers of English are intuitively aware
of the difference between related verbs such as those referring to loss of
possession (e.g. steal, purloin, pilfer, filch, swipe) or those referring to the
emission of a loud sound (e.g. scream, shriek, screech, bawl).
However, the ability of language users to differentiate between related
predicate meanings not only entails the consideration of single events (e.g.
drink) as compared to others in the same lexical domain (tipple, sip), but also
an awareness of the configuration of subevents designated by the same verb
within a larger ‘frame’ (Fillmore 1977, 1982). Frame Semantics (Fillmore
2006) studies how linguistic forms evoke frame knowledge, and how the
frames thus activated can be integrated into an understanding of the pas-
sages that contain these forms. Frames are crucial elements to consider in
the representation of lexical meaning.
For example, in certain cases, frames override constraints on the events
that a single predicate can encode. According to Croft (1991, 2012), the only
way in which two subevents can co-exist in the same verb is if they are
causally related (e.g. smash, which encodes the action of smashing as well as
its result). However, this is not necessarily true. Goldberg (2010: 44–49) states
that the verb double-cross designates an event of betrayal following a state or
event of understood cooperation. The betrayal is not caused by the state of
trust, nor does the betrayal cause the state of trust. Instead the state of trust
is part of the background frame that is presupposed in order for the profiled
or asserted act to count as double-crossing. This one subevent is profiled,
while the background frame presupposes one or more subevents without a
causal relation between them. Frame knowledge is thus essential to under-
standing the meaning of the verb.
A considerable part of a speaker’s language competence is the ability to
choose the right predicate and to select the most suitable arguments to fill
the slots activated by it in the communicative context. An effective lexical
representation system should be capable of accounting for differences in
lexical meaning within the context of a semantic domain and frame.
The first step in designing a representation for this purpose would
presumably involve the specification of a core set of basic concepts that
serves as a foundation for meaning representations. This is the only way to
obtain a formal representation with sufficient explanatory adequacy to
account for those aspects of meaning that transcend the code itself and
are part of the shared cultural knowledge and frames of a given speaker
community.
Since Foley and Van Valin (1984), the core component of RRG has been the
bidirectional linking algorithm that captures how syntactic and semantic
representations are mapped onto each other. This linking algorithm com-
prises a semantic level and a syntactic level, which model the communication
act between a speaker and an addressee. The semantic phase of the linking
algorithm begins in the lexicon, where a meaning representation is con-
structed from the information stored there. This representation is the input
for the syntactic phase, which involves the assignment of syntactic functions
and morphosyntactic properties. The syntactic phase, which proceeds in the
opposite direction, starts with an utterance. After the application of a syntac-
tic parser, the morphosyntactic properties of the input sentence are repre-
sented by means of the layered structure of the clause (LSC). This syntactic
representation is a combination of both universal and language-specific dis-
tinctions. The interpretation of the sentence stems from mapping rules that
link the syntactic representation to the semantic representation.
Adjectives can also occur as part of attributive predications. Note that be′ is
only found in the representation of attributive predications, not in the
representation of the resulting state; moreover, be′ is not the representation
of the copula, but the operator of a state, comparable to the operator do′.
The operator be′ thus figures in these representations as the principal
characteristic of attributive logical structures:
(2) a. Laura is wealthy.
be′ (Laura, [wealthy′])
b. John is afraid of birds.
feel′ (John, [afraid′ (birds)])
In regard to adverbs, Van Valin (2005: 49) explains that they are repre-
sented as one-place predicates and can modify different parts of a logical
structure. For example, the argument of a time adverb can be the entire
logical structure. In (3), the entire event of Lucas walking and then beginning
to be at the zoo takes place within the time frame designated by tomorrow.
In (6), a sustained activity is performed by the juror, who does not stop
moving throughout the time period of the stated event (trial).
Finally, Van Valin (2005) describes how predicative prepositions are repre-
sented and underlines the fact that they are also state predicates. As shown
in (7), the logical structure be-in′ is the highest predicate since the grassland
is the place where the kangaroo is jumping. As such, it has two arguments,
namely, the grassland and the logical structure for jump.
This contrasts with the logical structure of the active accomplishment, the
kangaroo jumped to the grassland, in which the predicative preposition (PP)
indicates the final location of the kangaroo and not the location of the
jumping event.
Although, unlike verbs, nouns (e.g. cat and table) lack a logical structure, they
have semantic properties in the form of qualia roles, which parameterize
their meaning. According to Pustejovsky et al. (2006: 3), nouns can be
described in terms of the following:
coffee knife book die eat read red heavy dangerous good rising frightened
Figure 3.2 Qualia structure of violin (Pustejovsky and Batiukova 2019: 162)
predicates (love′ and snore′) are used in their own definitions. Nonetheless,
love, which is a verb of feeling, could be defined in terms of feel, whereas snore
is a type of sound emission.
The same also occurs in (10) and (11) except for the fact that the BECOME
operator introduces the resulting state.5
(14) show [do′ (x, Ø)] CAUSE [BECOME see′ (y, z)]
According to Van Valin (1993) and Van Valin and Mairal Usón (2014), it is
necessary to improve the lexical representation system so as to obtain more
consistent semantic decompositions. However, part of the problem lies in
the fact that logical structures, as they were initially conceived, only focus
on capturing grammatically relevant aspects of word meaning and neglect
contextual meaning.
an ontology (in the artificial intelligence sense rather than the philosophical
sense), defined by Gruber (1976) as an explicit specification of a conceptual-
ization. One crucial property of a conceptual system is that no concept can
be described without an account of its relationships to others (Lamb
1998: 147).
When such representations are anchored to a well-designed conceptual
ontology or network that relates word senses, the primitives in a conceptual
lexical structure are no longer regarded as predicates but rather as concep-
tual units taken directly from an ontology.6
The ontology that can be used to specify the concepts in conceptual logical
structures is FunGramKB, a multilingual online environment for the semi-
automatic construction of a multipurpose lexico-conceptual knowledge
base for natural language processing (NLP) systems (Periñan-Pascual and
Arcas-Tuñez 2007, 2010; Mairal Usón and Periñan-Pascual 2009).7 The struc-
ture of FunGramKB has two levels: a linguistic level and a conceptual level.
The three components of the linguistic level are the lexicon, grammaticon
and morphicon, which store the lexical, grammatical and morphological
information for each word. In a parallel way, the conceptual level is also
composed of three modules: the ontology (concept hierarchy), the cognicon
(scripts encoding procedural knowledge) and the onomasticon (proper
names of entities and events).
In the same way as other lexically based ontologies, the main premise of
FunGramKB is that the conceptual representations of objects, attributes and
events are mapped onto language in some significant way. The assumption
is that the structure of the lexicon is based on a core set of undefinables or
conceptual invariants and that these invariants can be extracted from
lexicographic resources that document our shared knowledge of the world.
Whereas the lexicon houses language-specific syntactic and morphological
information, the language-neutral ontology captures the meanings and
interrelationships of concepts that are not explicitly stated in the lexicon.
The first step in building the ontology was the specification of an inven-
tory of basic concepts that are assumed to be lexicalized in a wide range of
languages. The ontology acquisition methodology is based on the extraction
of type_of hierarchies from dictionary definitions (Martín-Mingorance 1984,
1990, 1995; Hirst 2009; Amsler 1980, 1981). Whereas the genus designates
the superordinate concept of the defined word, the differentiating features
are the properties that make the concept different from other members of
the same conceptual category. The meaning of a word is thus an access point
to a concept or conceptual structure of some kind. Pustejovsky (2001: 5) also
claims that his qualia are linked to a network of concepts (though this
relational structure is never specified). Consequently, all problems are trans-
ferred to the ontology since each word sense is represented as a pointer to
some concept or category within the ontology.
The FunGramKB ontology has the following levels. At the highest level
are meta-concepts (e.g. #EVENT). At the next level are basic concepts
(e.g. þHUMAN_00), which are used to define other basic concepts and
terminal concepts. At still another level are terminal concepts (e.g.
$METEORITE_00), which are more specific and not used in the meaning
postulates of other concepts. The final level of subconcepts (e.g.
PASTEURIZE), has the same thematic frame as a terminal concept (i.e.
$STERILIZE_00), but specifies one of its arguments to a greater degree. (In
the case of PASTEURIZE, the second argument is restricted to milk prod-
ucts.) The concepts at the meta-conceptual, basic, terminal, and subconcept
levels are preceded by the symbols, #, þ, $, and , respectively. Except for
the subconcept level, each also has a numerical code, (_00, _01, _002, etc.).
The methodology used to extract concepts consists of four phases: (1) con-
ceptualization, (2) hierarchization, (3) remodelling, and (4) refinement.8 The
source was the Longman Dictionary of Contemporary English (Procter 1978). All
basic and terminal concepts in FunGramKB are described in terms of seman-
tic properties represented in thematic frames and meaning postulates. Each
event is assigned to one thematic frame.
A thematic frame contains the number and type of participants (with
selection restrictions) involved in the event or quality. Although not all
participants are lexicalized, they are implicit to the understanding of the
concept. The second type of conceptual schema is a meaning postulate,
which is a set of one or more logically connected predications (e1, e2, etc.)
with the generic features of concepts. Only one thematic frame and one
meaning postulate are assigned to each concept.
Meaning postulates are encoded in COREL (Conceptual Representation
Language) (Periñán-Pascual and Mairal Usón 2010). This machine-readable
language defines all of the units. It can also be used to obtain conceptual
relations by applying inheritance and inference mechanisms to the mean-
ing postulates (Periñán-Pascual and Arcas-Túnez 2005). In each representa-
tion, the meaning postulates provide the basis for natural semantic
decomposition since each lexeme is linked to a concept.
Moreover, the fact that lexical entries are linked to conceptual units and
thus have different conceptual routes is a way of resolving polysemy. CLSs
facilitate the selection of the most suitable concept in the case of predicates
with more than one meaning. In this sense, each meaning of a predicate is
linked to a different concept. The result is a robust enumerative lexicon
with fine-grained meaning postulates, capable of generating a complex
conceptual network. It is thus no longer necessary to condense all the
meanings of a lexical unit into a single infra-specified representation as
proposed by Generative Lexicon Theory.
In CLSs, differences in meaning are not marked by the syntactic behaviour
of a lexeme but rather are specified in its meaning. Accordingly, nuances of
meaning resulting from differences in syntactic constructions do not affect
the conceptual representation or the selection of the conceptual unit. This
method of dealing with polysemy is exemplified by fix. For instance, one of
the various meanings of fix is to repair something that is broken or not
working properly (15).
The selection of (17a) or (17b) depends on the basic conceptual event since
cooking is a creation event whereas repairing is a change event. Accordingly,
þREPAIR_00 and þCOOK_00 differ in their conceptual routes and positions
in the ontology. In the case of þCOOK_00, the conceptual path is
#MATERIAL >> þDO_00 >> þCREATE_00 >> þCOOK. Its meaning postu-
late is the following:
(18) þ(e1: þCREATE_00 (x1) Theme (x2) Referent (f1: þHUMAN_00) Beneficiary
(f2) Instrument (f3: þFOOD_00)Means (f4: (e2: þHEAT_00 (x1) Theme (x3: f3)
Referent))Manner)
This CLS for obsolete consists of the conceptual unit $OBSOLETE_00, which,
in turn, is composed of a thematic frame and a meaning postulate
(Figure 3.6). Since it is a state, the thematic frame only has one participant
(x1), which is a previously useful entity that is now socially perceived
as useless.
Obsolete is a predicate that can be applied to a wide variety of objects, such
as typewriter. As is well known, a typewriter is a machine with keys that are
pressed to print letters of the alphabet onto paper. Since the 1990s, personal
computers more efficiently perform the functions of typewriters. Thus,
understanding why typewriters have become obsolete entails elements of
world knowledge, such as an awareness of past and present time frames as
well as of the affordances of typewriters in comparison to those of more
modern machines that can be used for the same purpose. The meaning
postulate in Figure 3.6 describes the three events that occur when an entity,
such as a typewriter, is described as obsolete:
As a noun, invoice does not have a thematic frame. However, this does not
mean that it lacks a relational structure. As a commercial document that
itemizes a transaction between a buyer and a seller, invoice indirectly refers
to a complex knowledge configuration, which speakers have to access in
order to understand the term. This information should be included in the
meaning postulate of the conceptual entry. Accordingly, the conceptual
entry for þINVOICE_00 in Figure 3.8 is composed of a meaning postulate
with two main events, which state what type of entity an invoice is (e1) and
the context in which it is produced (e2).
This second event contains three subevents (e3, e4, e5). Examples (28–31)
show each event and subevent within the corresponding frame, which
encodes the purpose of an invoice (f1) and the possible reasons for issuing
it (f2, f3).
(30) (f1: (e3: þPAY_00 (x4) Agent (x5)Theme (x6) Origin (x3) Goal))Purpose
[The purpose of an invoice is for x4 to pay x3.]
3.4 Conclusion
RRG acknowledges the need for a more detailed theory of semantic repre-
sentation in order to justify semantic decomposition as well as the primitive
states and activities upon which it is based. This requires the formulation of
a representation that is not limited to syntactically relevant factors. Such a
representation system would be able to specify which lexeme would serve as
the generic term in the definition of a concept and would allow users to
effectively retrieve the semantic properties of each predicate in the lexicon.
RRG requires an ontological semantic theory, which is the natural evolution
of previous efforts to improve lexical representations, such as lexical tem-
plates with internal and external variables, based on lexical functions
and qualia.
For this reason, this chapter has provided a detailed description of a
conceptually oriented representation system that accounts for the non-
propositional dimension of meaning, and which allows users to access
contextual or encyclopedic meaning. In addition, it argues that adopting
an ontological approach in the form of CLSs instead of standard LSs in RRG
has numerous advantages. For example, CLSs can be automatically trans-
duced to a COREL scheme which can be used to obtain inference operations
with a reasoning engine. Consequently, the addition of CLSs to the RRG
system substantially enriches its semantic representations.
References
Notes
4.1 Introduction
Rappaport Hovav and Levin (1988), Grimshaw (1990) and Tenny (1994)
advocated an aspectual approach. According to these authors, the aspectual
properties of a predicate alone determine the syntactic realization of the
arguments. Telicity was deemed to be a pertinent criterion: an argument
affected by a telic process has to be realized as an object at some level of
representation.3 However, Rappaport Hovav and Levin (2015: 606) pointed out
that change-of-state verbs, the most stable class in terms of their linking
properties, are aspectually heterogeneous, including telic or variably telic
verbs (degree achievements according to Dowty (1979), e.g. flatten versus
widen).4 Moreover, no aspectual criterion is available to predict the linking
of two-place state predicates such as those in (3) and (4).
Hence, not only aspectual properties, but also non-aspectual properties such
as sensation, perception, volition and control, play an important role in argu-
ment linking. According to François (1997), aspectual as well as participant-
oriented properties (sensation, perception, control and causation5) are indis-
pensable for argument linking. Macroroles in RRG as instantiations of general-
ized semantic roles are a construct that allows such an integration.
Table 4.1 Logical structures of states and activities and thematic relations (see Van
Valin and LaPolla 1997: 115)
I. STATE VERBS
A. Single argument
1. State or condition broken′ (x) x ¼ PATIENT
2. Existence exist′ (x) x ¼ ENTITY
B. Two arguments
1. Pure location be-LOC′ (x, y) x ¼ LOCATION , y ¼ THEME
2. Perception hear′ (x, y) x ¼ PERCEIVER , y ¼ STIMULUS
3. Cognition know′ (x, y) x ¼ COGNIZER , y ¼ CONTENT
4. Desire want′ (x, y) x ¼ WANTER , y ¼ DESIRE
5. Propositional att. consider′ (x, y) x ¼ JUDGER , y ¼ JUDGEMENT
6. Possession have′ (x, y) x ¼ POSSESSOR , y ¼ POSSESSED
7. Internal experience feel′ (x, y) x ¼ EXPERIENCER , y ¼ SENSATION
8. Emotion love′ (x, y) x ¼ EMOTER , y ¼ TARGET
9. Attrib./Identific. be′ (x, y) x ¼ ATTRIBUTANT , y ¼ ATTRIBUTE
Table 4.2 Actor-Undergoer Hierarchy (see Van Valin and LaPolla 1997: 127, 146;
Van Valin 2005: 61)
A CTOR U NDERGOER
Two-place states (pred′ (x, y)) formalize the semantics of bivalent stative
verbs of possession or emotion such as own or like. In line with localist
approaches (Gruber [1965] 1976; Jackendoff 1972), the x-argument of (pred′
x, y) could be thought of informally as a specific instance of a location. The
y-argument of (pred′ x, y) informally corresponds to specific instances of a
theme in terms of Gruber ([1965] 1976).
To sum up, the AUH represents a continuum of argument positions in
different subparts of lexical representations. The leftmost argument is the
prototypical candidate for actor, the rightmost argument the prototypical
candidate for undergoer.11 By default, the number of macroroles follows
from the LS. For lexical representations with no arguments, for example
rain′, no macrorole will be assigned. One-place intransitive activities, for
example do′ (x, [shine′ (x)]), will take an actor. One-place intransitive non-
activities, for example exist′ (x), will take an undergoer. Macroroles always
correspond to direct core arguments, although not every direct core argu-
ment bears a macrorole.12 By default, transitive predicates with two or three
arguments take two macroroles. The default macrorole assignment
principles are the following.
(6) a. The hunter [A(ctor)] killed the [do′ (x, Ø]CAUSE[BECOME dead′ (y)].13
bear [U(ndergoer)]
b. The child [A] saw her friends [U] do′ (x, [see′ (x, y)])
c. The girl [A] was running do′ (x, [run′ (x)])
d. The stars [A] were shining do′ (x, [shine′ (x)])
e. The ghost [U] did not exist exist′ (x)
For transitive verbs, the default macrorole assignment rules predict the
number of macroroles but do not stipulate which argument takes the actor
macrorole and which argument takes the undergoer macrorole. The default
choice follows from the consideration that ‘given the LS for a transitive verb,
the leftmost argument in it will be the actor and the rightmost will be the
undergoer’ (Van Valin 2013: 78).
It follows from the examples in Van Valin (2013: 78) that this also holds for
causatives with two LS components (cause and effect) linked together with a
CAUSE operator. Starting from the left with the αCAUSE component of the
LS, in example (8) the actor macrorole is assigned to the x-argument (dog).
Undergoer assignment will start from the right of the CAUSEβ component.
The rightmost accessible argument will be the y-argument (boy) that will be
selected for undergoer.
(8) The dog scared the boy. [do′ (x ¼ dog, Ø] CAUSE [feel′ (y ¼ boy, afraid′])]).
Note that if the undergoer were always the rightmost argument, predicates
represented by a causative structure with a two-place state in the CAUSEβ
component should always select the second argument of pred′ (x, y) as
undergoer.
(9) give [do′ (x, Ø)] CAUSE [BECOME have′ (y, z)]
(10) load [do′ (x, Ø)] CAUSE [BECOME be-on′ (y, z)]
Van Valin (2005: 126) acknowledged that from a strictly typological view-
point, the markedness hypothesis has to be abandoned due to the fact that
in primary object languages15 the pred′ (x,. . .)-argument seems to be the
unmarked choice for undergoer (see Guerrero and Van Valin 2004; Diedrich-
sen 2008). While most languages follow principle A in (12), some languages
follow principle B.
(12) Principles for Undergoer-choice (Van Valin 2005: 126)
1. Principle A: lowest-ranking argument in LS (default)
2. Principle B: second-highest-ranking argument in LS
In this section, we deal with some types of predicates that can be said to be
challenging for the RRG theory of macrorole assignment.
4.4.1 Causativity
When Foley and Van Valin (1984) first proposed the AUH, they presupposed –
in line with Dowty (1979) – a logical link between causative constructions
and accomplishments. In addition, they considered the argument of
αCAUSE to be an argument of DO. As the argument of αCAUSE is necessarily
the most active argument in the hierarchy, no problem arose with macro-
role assignment.
(13) Joan broke the glass [DO (Joan)] CAUSE [BECOME broken′ (glass)]
(Foley and Van Valin 1984: 52)
However, later research in RRG (Van Valin and LaPolla 1997: 97, 106–109;
Van Valin 2018) has shown that equating causative constructions with
accomplishments was erroneous. On the one hand, there are accomplish-
ments that are not causative, like the one in (14a), which contrasts with its
causative counterpart in (14b).
Actor Undergoer
propose the LS [have′ (Bill, gun)] CAUSE [feel′ (Martha, [afraid′])] for (16a).
Both compounds of CAUSE consist of a two-place state predicate (have′(x ,y)
and feel′ (x, y)). For the β part the y-argument is a predicate [afraid′]. The two
x-arguments of these predicates (Bill and Martha) are both instances of the
same middle position of the AUH. If we follow the rule of processing from
left to right, Bill will be the actor. However, this is inaccurate from a
semantic point of view. It is not Bill, but his possession of a gun that is
responsible for causing Martha’s fear. Hence, the actor of frighten is not Bill,
but the whole αCAUSE compound. It may seem strange to think of a state of
affairs as an actor, but otherwise frighten would cease to be a macrorole
transitive predicate in (16a).
The binding facts seem to be the reason why the β component appears as
the y-argument of the two-place state and the α component as the argument
of the preposition co-referring with the y-argument. In a footnote Van Valin
and LaPolla (1997: 668) remark that about herself is an argument PP and not
an argument-adjunct, because the PP is ‘the optional realization of the α
variable’. The fact that the preposition seems to be semantically empty and
commutable (talk about, talk of as well as German sprechen über, sprechen von)
corroborates this view. However, it is precisely for this reason that it seems
problematic to use an argument-adjunct construction to represent the third
argument of talk.20
Staudinger et al. (2008) proposed an alternative analysis. RRG allows for
several classes of two-place activities: do′ (x, [sing′ (x,(y))]); do′ (x, [see′ (x,(y))]);
do′ (x, [tap′ (x,(y))]), etc. (Van Valin and LaPolla 1997: 115).21 The second
argument represents an instance of pred′ (x, y) (ibid.: 127), but the predicate
embedded in activities is not a state. Hence, RRG could theoretically allow
for three-place activities with an argument in each of the three medium
positions of the AUH: do′ (x), pred′ (x, . . .), and pred′ (. . ., y). Staudinger et al.
(2008) thus suggested the following LS for French parler, the counterpart of
English talk.
Note that according to the formalism used in RRG, the predicate talk′
(x,(y),(z)) is not a state, similarly to sing′ (x, (y)) in do′ (x, [sing′ (x, (y))]). As
indicated before, activities do not consist of states embedded under a do′
operator. Hence, in an RRG description of talk, one could associate the two
arguments y and z with the positions pred′ (x, . . .) and pred′ (. . ., y) of the
AUH, as RRG associates the y-argument of sing′ with the pred′ (. . ., y)
position.
However, this solution would not be available for three-place non-
causative states. Verbs of envy are possible candidates for this class
(Kailuweit 2012: 118). In line with two-place stative verbs of emotion such
as love or hate, envy takes an emoter argument, the person who experiences
envy, and a target argument, the envied person. In addition, envy possesses
a third argument expressing the object or circumstance the target argu-
ment is envied for.22 For the English verb envy, this third argument could be
realized as a direct core argument (21a) or as an oblique (21b).
The fact that the third argument, the object of envy, can also be marked
by for (21b) may suggests that it is an adjunct, which gets incorporated into
the core as a direct argument, by analogy with recipient beneficiaries,
such as Max baked a cake for Paul vs. Max baked Paul a cake. Thus, a possible LS for
(21b) could be because.of′ ([have′ (Paul, wealth)], [envy′ (Max, Paul)]). However,
in (21a) and (21b) the target, the envied person, is the undergoer, while the
for-complement becomes the undergoer in the recipient-beneficiary
construction Max baked a cake for Paul ! Max baked Paul a cake.
French envier (‘envy’) shows variable undergoer choice. The target can be
realized either as an undergoer (22a) or as non-macrorole direct core argu-
ment in the dative case (22b). As in English (21c), the object of envy
argument is optional if the target argument is realized as undergoer.
realized as a direct core argument in the dative case and as an oblique object
respectively. In comparison with verbs of envy, en vouloir raises an additional
problem: the nature of the macrorole. RRG describes non-episodic verbs of
anger as instances of verbs of internal experience with the LS feel′ (x, y). The
x-argument corresponds to the experiencer, and the y-argument is a
predicate of sensation having one internal argument that is not accessible
for macrorole assignment. Therefore, the predicates of internal sensation
are by default M-intransitive. As intransitive states, they select an undergoer.
The undergoer macrorole is assigned to the x-argument, the experiencer.
(23) Episodic verbs of emotion (Van Valin and LaPolla 1997: 156, 402)
Pat is angry at Kelly.
[feel′ (Pat ¼ experiencer ! 1. arg. of pred′ (x, y) ! Undergoer ! PSA,
[angry.at′ (Kelly)] ¼ sensation ! 2. arg. of. pred′ (x, y))]
The fact that languages such as German or Russian possess direct core
arguments in the dative case, which behave differently from other non-
macrorole arguments in syntax, led to intense discussion of the necessity
for a third macrorole. The German passive construction with bekommen
(‘get’) seems to be a candidate for illustrating the theoretical contribution
that a third macrorole could make.
Diedrichsen (2008) discusses the following data.
(24) LS: [do′ (Eltern, Ø)] CAUSE [BECOME have′ (ich, Computer)]
a. Active voice
Meine Eltern haben mir diesen Computer geschenkt
my parents have.3pl 1sg.dat dem.acc computer given
‘My parents gave me this computer.’
b. Passive voice
Dieser Computer ist mir von meinen Eltern geschenkt worden
dem.nom computer be.3sg 1sg.dat by my parents given become
‘This computer was given to me by my parents.’
c. Bekommen-passive voice
Ich habe diesen Computer von meinen Eltern geschenkt bekommen
I.nom have.1sg dem.acc computer by my parents given received
‘I got/received this computer, given by my parents’, ‘I was given this computer
by my parents.’
Figure 4.2 Four types of languages, accusative and ergative alignment (Haspelmath 2008)
different forms of alignment. For example, marking S and A in the same way
characterizes accusative alignment. While grouping O/P and X/T together
corresponds to indirective alignment, secundative alignment is character-
ized by marking O/P and G/R identically (Bickel 2011).
Haspelmath (2008) pursues a very similar approach, which he includes
directly into the RRG framework. He proposes four macroroles: A ¼ actor,
U ¼ undergoer of monotransitive sentences, R ¼ macro-recipient, corres-
ponding to Primus’s (1999) proto-recipient and the aforementioned indir-
ectus of Lehmann et al. (2004), and finally T ¼ (macro-)theme (see
Figure 4.2). These macroroles are defined in purely semantic terms. They
account for the parallels between monotransitive and ditransitive
alignments in a straightforward way without needing to refer to marked
or unmarked undergoer choice, as in standard RRG, or to parameterize
undergoer selection for the indirective–secundative contrast, as in Guer-
rero and Van Valin (2004).
Although U, T, R and A seem to correspond to the four rightmost pos-
itions of the AUH, it is unclear how their semantic values could be deter-
mined in a precise way. Haspelmath (2011) himself criticizes the approach
of Bickel et al. (2014), but his critique refers indirectly to his own RRG
approach, too (Haspelmath, p.c., 30 January 2013). His point is that Bickel’s
generalized semantic roles are universal semantic categories aimed at the
description of individual languages. However, being not only syntactically
undetermined but also semantically highly heterogeneous, they do not
serve to draw typological generalizations. Indeed, it remains unclear in
both approaches what agent-like, goal-like or theme-like mean beyond
purely intuitive notions.29
Table 4.3 Features and values for activity clusters (from Kailuweit 2013)
þc c c þm m m r r þr
4 2 0 2 1 0 0 2 4
In line with Comrie’s approach, Haspelmath (2011) points out that S, A, P/O/
U, T/X, and R/G should only be comparative concepts referring to the syntactic
relations associated in a given language with verbs selecting a prototypical
agent and patient (A:P), verbs of transfer (A:R:T), and one-place verbs of
uncontrolled change of state such as verbs of dying (S). This means that not
every two-place predicate assigns an active and a passive generalized semantic
role. Only predicates that code the opposition of the arguments, like proto-
typical verbs with an agent and a patient, do: for example verbs like English
see, like, own or enter follow the pattern of break or kill, but appeal to and belong
to do not. This is in line with standard RRG. However, Haspelmath’s (2011)
comparative concepts do not apply to all predicates in every language and do
not formulate hypotheses about universal semantic structure. Hence, they
differ considerably from RRG macroroles.
Kailuweit (2013, 2018) maintains the two macroroles of standard RRG,
devising, however, a feature-based Activity Hierarchy, which departs from
the RRG AUH. Following Rozwadowska (1988), Kailuweit (2013) takes three
features into account: causative and/or control [c], mental (sentient) [m],
and resultative (change of state) [r]. In line with Reinhart (2002), these
features can assume one of the three values þ, or . In addition, the
features are weighted (see Table 4.3). The feature [c] is a strong actor
feature, [m] is a weak actor feature, and [r] is a strong undergoer feature.
The presence of a strong feature, indicated with [þ], will duplicate the
value of the presence of a weak feature. If an argument is underspecified
for one feature or has an intermediate degree of that feature [], the value
will be half of the [þ] value.
The combination [þcþmr] represents the prototypical actor with the
value 4þ2þ0 ¼ 6, while the prototypical undergoer corresponds to the
combination [cmþr] 0þ04 ¼ 4. Nine intermediate combinations are
mathematically possible. Hence, the system distinguishes eleven possible
degrees of activity. The attribution of the values follows from a detailed
semantic analysis of events and participants, evaluating aspectual proper-
ties as well as causativity, sentience, responsibility and control.
In the following, sketch-like descriptions will be given to account for the
non-causative three-place predicates talk and envy as well as for some two-
place verbs of emotion. Some of the examples discussed above are repeated
for convenience.
4.6 Conclusion
References
Borer, Hagit. 2005. Structuring Sense II: The Normal Course of Events. Oxford:
Oxford University Press.
Comrie, Bernard. 1978. Ergativity. In Winfred P. Lehmann (ed.), Syntactic
Typology: Studies in the Phenomenology of Language, 329–394. Austin: Univer-
sity of Texas Press.
Diedrichsen, Elke. 2008. The grammaticalization of the bekommen-passive in
a RRG-perspective. In Kailuweit et al. (eds.), 87–145.
Dixon, Robert M. W. 1979. Ergativity. Language 55: 59–138.
Dowty, David. 1979. Word Meaning and Montague Grammar: The Semantics of Verbs
and Times in Generative Semantics and in Montague’s PTQ. Dordrecht: Reidel.
Dowty, David. 1991. Thematic proto-roles and argument selection. Language
67(3): 547–619.
Dryer, Matthew S. 1986. Primary objects, secondary objects, and antidative.
Language 62: 808–845.
Fillmore, Charles J. 1968. The case for case. In Emmon Bach and Robert
Harms (eds.), Universals in Linguistic Theory, 1–88. New York: Holt, Rinehart
and Winston.
Fillmore, Charles J. 2003. Valency and semantic roles: The concept of deep
structure case. In Vilmos Ágel, Ludwig M. Eichinger, Hans Werner Eroms,
Peter Hellwig, Hans Jürgen Heringer and Henning Lobin (eds.), Dependenz
und Valenz: Ein internationales Handbuch der zeitgenössischen Forschung /
Dependency and Valency: An International Handbook of Contemporary Research
1, 457–475. Berlin: Walter de Gruyter.
Foley, William A. and Robert D. Van Valin, Jr. 1984. Functional Syntax and
Universal Grammar. Cambridge: Cambridge University Press.
François, Jacques. 1997. La place de l’aspect et de la participation dans les
classements conceptuels des prédications verbales. In Jacques François and
Guy Denhière (eds.), Sémantique linguistique et psychologie cognitive. Aspects
théoriques et expérimentaux, 119–156. Grenoble: Presse Universitaire.
Grimshaw, Jane. 1990. Argument Structure. Cambridge, MA: MIT Press.
Gruber, Jeffrey S. 1965 [1976]. Studies in Lexical Relations. PhD dissertation,
MIT. Published as: Lexical Structures in Syntax and Semantics. Amsterdam:
North-Holland.
Guerrero, Lilián and Robert D. Van Valin, Jr. 2004. Yaqui and the analysis of
primary object Languages. International Journal of American Linguistics 70(3):
290–319.
Haspelmath, Martin. 2008. Ditransitive constructions: Towards a new Role
and Reference Grammar account? In Van Valin (ed.), 75–100.
Haspelmath, Martin. 2011. On S, A, P, T, and R as comparative concepts for
alignment typology. Linguistic Typology 15: 535–689.
Hockett, Charles. 1958. A Course in Modern Linguistics. New York: Macmillan.
Ibáñez Cerda, Sergio. 2008. ‘Saying’ verbs in Spanish. Deepening the lexical
semantics description. In Van Valin (ed.), 3–21.
Jackendoff, Ray. 1972. Semantic Interpretation in Generative Grammar. Cam-
bridge, MA: MIT Press.
Nissenbaum, Helen Fay. 1985. Emotion and Focus. Stanford, CA: CSLI
Publications.
Pinker, Steven. 1989. Learnability and Cognition: The acquisition of Argument
Structure. Cambridge, MA: MIT Press.
Primus, Beatrice. 1999. Case and Thematic Roles: Ergative, Accusative, Active.
Tübingen: Niemeyer.
Rappaport Hovav, Malka and Beth Levin. 1988. What to do with theta roles.
In Wilkins (ed.), 7–36.
Rappaport Hovav, Malka and Beth Levin. 2015. The syntax–semantics inter-
face. In Shalom Lappin and Chris Fox (eds.), The Handbook of Contemporary
Semantic Theory (2nd ed.), 593–624. Oxford: Wiley-Blackwell.
Rauh, Gisa. 1988. Tiefenkasus, thematische Relationen und Thetarollen. Tübingen:
Narr.
Reinhart, Tanya. 2002. The Theta system: An overview. Theoretical Linguistics
28(3): 229–290.
Rozwadowska, Bozena. 1988. Thematic restrictions on derived nominals. In
Wilkins (ed.), 147–165.
Ruwet, Nicolas. 1972. Théorie syntaxique et syntaxe du français. Paris: Seuil.
Staudinger, Eva, Matthias Hartung and Rolf Kailuweit. 2008. Linking syntax
to semantics: template selection and PP-Attachment ambiguities. In Kai-
luweit, Wiemer, Staudinger and Matasović (eds.), 389–413.
Tenny, Carol. 1994. Aspectual Roles and the Syntax–Semantics Interface. Dor-
drecht: Kluwer.
Tesnière, Lucien. [1959] 1965. Eléments de syntaxe structurale. Paris:
Klincksiek.
Van Valin, Robert D., Jr. 1993. A synopsis of Role and Reference Grammar. In
Robert D. Van Valin Jr. (ed.), Advances in Role and Reference Grammar, 1–164.
Amsterdam: John Benjamins.
Van Valin, Robert D., Jr. 1999. Generalized semantic roles and the syntax–
semantics interface. In F. Corblin, C. Dobrovie-Sorin and J.-M. Marandin
(eds.), Empirical Issues in Formal Syntax and Semantics, Vol. 2, 373–389. The
Hague: Thesus.
Van Valin, Robert D., Jr. 2004. Semantic macroroles in Role and Reference
Grammar. In Rolf Kailuweit and Martin Hummel (eds.), Semantische Rollen,
62–82. Tübingen: Narr.
Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cam-
bridge: Cambridge University Press.
Van Valin, Robert D, Jr. 2006. Semantic macroroles and language processing.
In I. Bornkessel et al. (eds.), Semantic Role Universals and Argument Linking:
Theoretical, Typological and Psycholinguistic Perspectives, 263–302. Berlin: Mou-
ton de Gruyter.
Van Valin, Robert D., Jr. 2007. The Role and Reference Grammar analysis of
three-place predicates. Suvremena Lingvistika 63(1): 31–64.
Van Valin, Robert D., Jr. (ed.). 2008. Investigations of the Syntax–Semantics–
Pragmatics Interface. Amsterdam: John Benjamins.
Van Valin, Robert D., Jr. 2010. Role and Reference Grammar as a framework
for linguistic analysis. In Bernd Heine and Heiko Narrog (eds.), The Oxford
Handbook of Linguistic Analysis, 703–738. Oxford: Oxford University Press.
Van Valin, Robert D., Jr. 2013. Lexical representation, co-composition, and
linking syntax and semantics. In James Pustejovsky, Pierrette Bouillon,
Hitoshi Isahara, Kyoko Kanzaki and Chungmin Lee (eds.), Advances in
Generative Lexicon Theory, 67–107. Dordrecht: Springer.
Van Valin, Robert D., Jr. 2018. Some issues regarding (active) accomplish-
ments. In Rolf Kailuweit, Lisann Künkel and Eva Staudinger (eds.), Applying
and Expanding Role and Reference Grammar, 5–26. Freiburg: Freiburg Univer-
sity Library.
Van Valin, Robert D., Jr. and Randy R. LaPolla. 1997. Syntax: Structure, Meaning
and Function. Cambridge: Cambridge University Press.
Van Valin, Robert. D., Jr. and David Wilkins 1996. The case for ‘effector’: Case
roles, agents, and agency revisited. In Masayoshi Shibatani and Sarah. A.
Thompson (eds.), Grammatical Constructions: Their Form and Meaning,
289–321. Oxford: Oxford University Press.
Wilkins, Wendy (ed.). 1988. Thematic Relations, Vol. 21: Syntax and Semantics.
New York: Academic Press.
Notes
16 Van Valin (2004: 74–78) pointed out that macroroles are basically seman-
tic categories that play a role in syntax. Syntax is considered non-
autonomous in RRG. Therefore, syntactic (in)transitivity is at least partly
semantically motivated.
17 Primus (1999: 36) substitutes Dowty’s (1991) proto-agent property ‘vol-
ition’ by ‘control’ to describe cases that exclude ‘volition’ in the strict
sense, for example responsibility or the mere capacity to start or stop an
event. In RRG, too, responsibility is referred to, at least informally, to
define the properties of the actor: ‘the actor is the entity to which
responsibility for the action or event is attributed’ (Van Valin and LaPolla
1997: 143).
18 ‘Seeing, thinking, liking, believing, etc., involve some kind of internal
activity (mental, emotional or perceptual) on the part of the participant,
whereas being seen, being thought about, being liked or being believed
does not require any action or effort of any kind on the part of the
participant. Hence the participant denoted by the first argument is more
active and hence more agent-like than the participant referred to by the
second argument, and, accordingly, the first argument is closer to the
agent end of the hierarchy than the second argument’ (Van Valin
2005: 58).
19 The existence of non-causative three-place predicates is widely ignored in
the literature. Haspelmath (2008: 14) refers at least incidentally to talk as
a verb ‘lacking an undergoer’, adding that ‘intuitively it is strange to
claim that ditransitive constructions are intransitive’.
20 Ibáñez Cerda (2008: 6) criticizes this representation, highlighting the fact
that the α variable could be realized as a direct object with verbs of saying.
21 Two other classes of two-place activities are verbs of consumption and
creation with an unspecified object: eat pizza, write letters. Note that the
objects of these predicates are not affected undergoers.
22 ‘Envy defines three central roles, that of the envier, the envied, and a
feature or possession of the one envied, over which he is envied’ (Nissen-
baum 1985: 108).
23 The clitic en is a desemantized part of the lexical realization of the
predicate that could be interpreted as a generic placeholder for the state
of affairs (the correlate) the anger is about. Curiously, the de-comple-
ment that can normally be substituted by the clitic en appears alongside
the clitic in the sentence. It is obviously not a fourth argument, but a
coreferential concretization of the correlate.
24 Spanish rabiar is somewhat different. It does not only accept a person as
the second argument but also an object. Hence, it corresponds to English
rage against.
25 For a discussion of Primus’s approach, and its compatibility with the
RRG framework, see Bellosta von Colbe (2004).
26 Bellosta von Colbe (2004) uses the same argument to criticize Primus’s
proto-recipient approach.
27 Bickel et al. (2014) use P (for patient) instead of O and T (for theme)
instead of X. See Haspelmath (2011) for an overview on the history of
denominations for generalized roles in typology.
28 Haspelmath (2011) and other authors refer to G as R (for recipient).
29 Bickel et al. (2014) refer to entailments in line with Dowty (1991), but
Dowty’s approach allowing 1,024 theoretically possible combinations
of proto-role features turns out to be a rather inaccurate tool to deter-
mine the activity degree of an argument in comparison with the AUH.
30 Note that French envier and German neiden/beneiden allow for flexible
undergoer selection, i.e. a construction selecting the second passive
argument as undergoer (secondary alignment in a language of indirec-
tive alignment). Kailuweit’s (2013) approach does not account for
this alternation.
31 Kailuweit (2018) discusses further classes of predicates, especially verbs
of directed motion such as climb, enter or reach.
5.1 Introduction
This chapter discusses one of the major advances of Role and Reference
Grammar (RRG) relative to other theories of grammar: the view of grammat-
ical relations as construction-based, and so not only language-specific but
construction-specific, rather than being global categories of the whole lan-
guage and found in every language. We also discuss the RRG conception of
the function of grammatical relations in referent tracking, which was one of
the major insights that led to the development (and naming1) of RRG. These
two insights have influenced the development of ideas outside RRG.
5.2 Background
Before the mid 1970s, a common assumption among linguists was that there
is a global category in all languages called ‘subject’ as well as other gram-
matical relations that we can talk about, and most theories assumed some
conception of grammatical relations, though there was much disagreement
about and no universal notion of ‘subject’, the grammatical relation dis-
cussed the most (Platt 1971; Van Valin 1977, 1981; Foley and Van Valin 1977,
1984; Keenan 1976; Gary and Keenan 1977; Comrie 1981). The word ‘subject’
derives from a Latin translation (subiectum) of Greek hypokeímenon ‘the under-
lying thing’, a concept that began with Aristotle’s theory of truth, where
Aristotle defined ‘subject’ as the entity that can have a predication about it,
that is, what the proposition is about, the topic about which a predication is
made. Aristotle did not have a separate term for grammatical subject. This
led to centuries of debate about the nature of subject (see Seuren 1998,
pp. 120–133, for an overview). Attempts were made to distinguish grammat-
ical subject from psychological subject (e.g. von der Gabelentz 1869: 378),
the latter essentially topic, and what was called ‘theme’ in the Prague School
terminology. A third term, ‘logical subject’ (often now seen as agent), was
sometimes used, but different scholars associated it either with grammat-
ical subject or with psychological subject (particularly in logic). Bloomfield
(1914: 61) used the term ‘subject’ to refer to topics and also to heads
of phrases.
Starting with Van Valin (1977, 1981) and Foley and Van Valin (1977,
1984), there were challenges to the notion of ‘subject’ and other gram-
matical relations (‘direct object’, ‘indirect object’) as global categories
within a single language, and as valid categories cross-linguistically (see
also Dryer 1997). Currently there are three major positions on this ques-
tion: (1) grammatical relations are global within a language and universal
cross-linguistically, and just need to be identified in different languages
(the rationalist/generativist/Chomskyan tradition); (2) grammatical rela-
tions exist, but are not necessarily global and not universal, and so
need to be defined in each language in terms of the constructions that
manifest them, if there are any (most empiricist/typological/explanatory
approaches); and (3) there are no grammatical relations, only part–whole
relations within constructions (Radical Construction Grammar; Croft
2001, 2013). Marantz (1982, 1984) has argued that grammatical relations
should not be seen as primitives or tied to semantic roles. For example,
‘subject’, as a grammatical category, is not simply a particular semantic
role, such as agent (see also Jespersen 1909–1949, vol. III, 11.1). ‘Subject’ is
also not simply topic; it must have grammatical properties beyond just
being what the clause is about. Empiricist linguists would generally agree
with this position.
been seen as part of one grammatical category (e.g. ‘subject’), they are not
one category, but instead are individual ways of constraining the interpret-
ation of who is doing what to whom, and languages differ in terms of
whether or not they constrain this functional domain at all, and if they do
constrain it, they differ in terms of which particular roles are identified, and
the particular mechanisms used to constrain the interpretation.
(1) a. Bobcontroller handed Jim the money and (pivot) left. [A,S]
b. Jimcontroller was handed the money by Bob and (pivot) left. [S,S]
c. The moneycontroller was handed to Jim but (pivot) not seen again
after that. [S,S]
d. Jimcontroller took the money and (pivot) was seen later buying a
new car. [A,S]
e. Jimcontroller took the money and (pivot) thanked Bob for it. [A,A]
f. Jimcontroller smiled and (pivot) took the money. [S,A]
5.3.2 Alignment
The particular PSAs in the conjoined clause coreference construction dis-
cussed here are found in English, but many other languages, even closely
related ones, do not manifest PSAs in conjoined clause coreference construc-
tions, and so the determination of the relevant argument of the second
These examples parallel the English examples in (1), but the interpretation
of the implied argument in the second clause of (2a) is obligatorily corefer-
ential with the Undergoer in the first clause, that is, it must be that the dog
is the controller and pivot (i.e. is the one that went downhill). In order to
have coreference that involves an A argument, the Dyirbal antipassive
construction must be used. This construction is an intransitive construction
with the verb marked with the antipassive marker ŋa, and has the Actor as
the single direct argument in the absolutive case and the Undergoer in the
dative case. Because Dyirbal has this antipassive construction, there is vari-
able access to the controller and pivot positions, and so the controller is a
variable syntactic controller and the PSA is a variable syntactic PSA, though
they manifest a different restricted neutralization from the corresponding
English construction. These grammaticalized constraints on interpretation
we have been looking at force a particular interpretation of an utterance in
both English and Dyirbal, but as the restricted neutralizations are different,
the interpretations are different. For example, if The man saw the dog and went
downhill is said in English, the interpretation has to be that the man went
downhill; but if the corresponding structure is used in Dyirbal, as in (2c), the
meaning has to be that the dog went downhill. We see here that the
construction must be taken as a whole, as it is the total construction
that influences the interpretation, and is not simply the sum of the individ-
ual words.
Where there is a choice of argument for PSA, the RRG theory of PSA
selection posits a default choice, depending on the Privileged Syntactic
Yet it isn’t the case that all languages necessarily have such a default
choice. As Foley and Van Valin argued (1984, §4.3) Tagalog has many differ-
ent constructions for having different semantic roles as PSA, yet none is a
default choice. In all these languages where we have seen a choice of PSA,
the choice of pivot is determined by pragmatic factors, such as the identifia-
bility or topicality of the referent involved (the default is used when there is
no difference in terms of the pragmatic factors), and so we refer to these as
pragmatic PSAs, as opposed to those based strictly on semantic factors,
which we call semantic PSAs.
While the conjoined cross-clause coreference construction in English
manifests an [S,A] restricted neutralization, it is not the case that all con-
structions in English manifest a restricted neutralization, and it is not the
case that all of the constructions that do manifest a PSA in English manifest
the same restricted neutralization. For example, in the following example of
the English purposive construction the controller and pivot of the
construction are the Undergoers of the two clauses, not the Actor or S of
the clauses:
(most commonly ’yung) to mark it as the Topic. The set of pronouns, which
appear as second-position clitics, also distinguishes Topical from non-
Topical referents, with the latter appearing as possessive forms or dative
forms. In (7) there is a short natural conversation to exemplify this feature
(from my own fieldwork; see also Schachter (2008: 337–338) for sets of
constructed parallel examples with the same arguments but with different
choices of topic, and Schachter and Otanes (1972, Ch. 5) for many of the
major constructions used for marking different semantic roles).8
In (7), the first speaker, Jirehel, refers to the making of sauce using a form
(gumawa) where the infix -um- marks it as Actor-Topic (it happens that in this
utterance the speaker has made the relevant clause the Topic of a higher
clause, but the phenomenon is the same), then the second speaker, Wendy,
uses the same root in line 3 of the example, but in the irrealis Patient-Topic
construction, to profile the event from the point of view with the sauce as
the Topic. She then follows this in lines 4 and 5 with two tokens of the root
lagay ‘put’ in the Location-Topic construction, to keep the sauce as the Topic,
but with the sauce now having the semantic role of the location where the
tomatoes and salt and pepper are to be added. Notice how there is no overt
reference to the Topic in any of Wendy’s utterances (e.g .’yung salsa ‘the
sauce’ could have been added to the end of each of Wendy’s utterances in
lines 3–5, but it wasn’t), yet we can tell what is being talked about (what is
the pragmatic and grammatical Topic) because of the marking on
the predicate.
We now can return to the issue of grammatical relations, that is,
restricted neutralizations. The choice of different roles as Tagalog Topic
affects the particular form of the construction and the behaviour of the
Topic in this and other constructions, such as the Tagalog relative clause
construction, where the head of the construction generally must be the
Topic of the modifying predicate (i.e. the form of the predicate must be
the same as if it were a main clause with that referent as Topic). For
example, if we were to recast the clauses in (7) as relative clause construc-
tions (and with realis predicates), we would get the forms in (8):
From these examples we can see that the prefixes are used for the Actors of
transitive constructions (10a, b) and the single arguments of intransitive
constructions where the action is voluntary and so the argument is an Actor
(so-called SA), as in (10c), and the suffixes are used for the Undergoers of
transitive constructions and the single argument of intransitive construc-
tions where the action is involuntary (an Undergoer, so-called SP or SO), as
in (10d).9
coreferential or not, and there is marking on the predicate to show this (i.e.
whether the PSA is the same as in the following clause or different from the
following clause), as in the following Choctaw examples (adapted from
Heath 1977: 212):
As can be seen from these examples, even without overt arguments, the affix
on the predicate marking whether the PSA in the following clause is the
same or different from that of the marked clause constrains the interpret-
ation of who is doing what. The pivots in this sort of system are generally
invariable syntactic pivots.
We have seen above that there are different kinds of constructions lan-
guages can have for constraining the inference of who is doing what, if they
have any at all. They may have some constructions that have invariable
semantic pivots (i.e. restriction with no neutralization), such as the verbal
marking in Tagalog and Acehnese, or invariable syntactic pivots, such as
reference across conjoined clauses in Warlpiri and Enga, or variable syntac-
tic pivots, such as reference across clauses in Dyirbal and English, and the
relative clause construction in Tagalog. The restricted neutralization found
in a construction could treat [S,A] the same in opposition to [U], or it may
treat [S,U] the same in opposition to [A]. As this is a construction-specific
phenomenon, the same language may have different constructions with
different PSAs, as we saw with Tagalog (see also Van Valin 1981 on Jakaltek
and Van Valin and LaPolla 1997: 282ff. on Tzutujil), and constructions with
different PSAs can sometimes be combined into a single complex structure,
such as when a Tagalog reflexive construction (invariable semantic pivot)
and any type of Tagalog clause structure (variable syntactic pivot) are com-
bined to form a complex structure. And of course a language may not have
any constructions that manifest PSAs, such as Mandarin Chinese and Riau
Indonesian.
(13)
e phūŋí dɯ́ s ̀ ŋ dip bɯ́ à nɯ̀ ŋɯ̄ a:ʔmı̀
e e e
e phūŋ-í dɯ́ -s ̀ ŋ
e e dip bɯ́ -à
e nɯ̀ ŋɯ̄ -ap-ı̀
Apung-agt Adeu-loc hit pfv-tr.pst ps cry-tmdys-intr.pst
(i) ‘Apung hit Adeu and (Apung) cried’ or
(ii) ‘Apung hit Adeu and (Adeu) cried’, or
(iii) ‘Apung hit Adeu and (someone else) cried’
Utterances of this type are somewhat rare, though; more often only one
possible actor is mentioned, as in an utterance like John finished eating and
left, and so the conversational implicature that the actor is the same in both
clauses (and it is only an implicature at first) can become strengthened to
the point that it becomes conventionalized as the only possible interpret-
ation, as in English, where a clause such as John put the rock next to the
chameleon and turned brown has to mean that John turned brown, even if it
makes no sense, unlike in a language where this coreference pattern has not
conventionalized (e.g. Chinese) and so it would more likely be interpreted as
meaning the chameleon turned brown.
Agreement or cross-referencing on the verb develops as an unstressed
pronoun is reinforced by a stressed pronoun or full noun phrase often
We have seen that syntactic relations develop from a form that is repeated
over and over again in discourse to the point that it becomes conventional-
ized as an obligatory part of the language, and thereby forces a particular
interpretation where otherwise there would be two or more possible inter-
pretations. But why would speakers repeat a form so often that this would
happen?
The answer lies in the culture of the speakers of the language, their way of
thinking, their value system. For a form to be used often enough for it to
become conventionalized, it must constrain the interpretation process of
the addressee in a way that is important to the speaker, so important that
the speaker is willing to put extra effort into constraining the addressee’s
inferential process in that particular way to make it more likely the
addressee will ‘get it right’. That is, the speaker wants to make sure the
addressee will infer that part of the communicative intention correctly,
more so, possibly, than other parts of the intention, and often uses a
particular form that they have used successfully before (and/or other people
have used successfully before) to constrain the interpretation in the same
way, over and over again. (We are creatures of habit and imitation, and
although we sometimes innovate, we more often go along with our usual
habits and also will imitate others.)
In the case of syntactic relations, what must be important to the speakers
is that the addressee correctly infer the roles of the major participants. The
clearest example of this is the development of relation morphology on the
References
Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cam-
bridge: Cambridge University Press.
Van Valin, Robert D., Jr. 2009. Privileged syntactic arguments, pivots, and
controllers. In Lilián Guerrero Valenzuela, Sergio Ibáñez and Valeria
A. Belloro, (eds.), Studies in Role and Reference Grammar, 45–68. Mexico:
UNAM.
Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning
and Function. Cambridge: Cambridge University Press.
Notes
1 The name ‘Role and Reference Grammar’ derives from the early focus on
the semantic roles and pragmatic functions in discourse referent tracking
of grammatical relations.
2 Following best practice in typology, for language-specific (descriptive)
categories and constructions I will capitalize the initial letters of the
name of the category or construction, but for comparative concepts
I will not capitalize the first letter. So, for example ‘Actor’ refers to
the language-specific grammatical category manifesting a particular
neutralization of semantic roles in the language under discussion,
while ‘actor’ refers to the comparative concept of the one who per-
forms an action. As there are no universal or cross-linguistic gram-
matical categories, descriptive and comparative concepts need to be
kept distinct.
3 Note that the identification of the referent of it as the same as that of the
newspaper is not due to grammatical relations, but simply to inference;
there is nothing in the grammar that obligatorily constrains the inter-
pretation, the way the inference of the relationship between Jim and the
thrower of the newspaper is constrained by the grammar.
4 English S is itself a neutralization of semantic macroroles for grammat-
ical purposes that is not found in all languages (see below on Acehnese
and Tagalog). Actor and Undergoer are also language-specific restricted
neutralizations of semantic roles for grammatical purposes, hence are
called macroroles, but are at a lower level than the restricted neutraliza-
tions of macroroles we are talking about here. In English and many other
languages there is also variable access to Undergoer status when there is
both a theme and a recipient or location in the clause, e.g. in the con-
struction I gave the book to Mary, the book is the Undergoer, but in the
construction I gave Mary the book, Mary is the Undergoer, and in Load the
truck with hay, the truck is the Undergoer, while in the construction Load the
hay on the truck, the hay is the Undergoer. The choice of one construction or
the other usually depends on the relative topicality of the referents. See
Van Valin and LaPolla 1997: 144ff.
Abbreviations
6.1 Introduction
* This chapter has benefited significantly from comments by a peer reviewer and suggestions by Delia Bentley. Of course, any
remaining mistakes and obscurities are my own fault.
Consider the verb give in English or its equivalent in some other language. It
evokes a scene that includes a giver, something given, and a person or object
that receives what is given. As part of the meaning of the verb, these
arguments are included in the lexical entry for give but the arguments
may be realized in different ways syntactically. For example, argument
structure alternations discussed in the literature on English syntax include
examples such as the following.
In each of these examples, the sentence entails that someone gave, some-
thing was given, and someone was the recipient or intended recipient (see
Williams 2015: 199–202). The argument structure is shaped by the passive
constructions in (3a) and (3b). The examples in (3c) and (3d) show that each
may have one unspecified argument but, unlike many languages, without
any morphology to signal the change. The syntactic arrangement of the verb
and its arguments is significantly different in each case.
The linking of arguments has been a key element of RRG since its incep-
tion (Foley and Van Valin 1984). As pointed out by Van Valin and LaPolla
(1997: 384, 389–392), the linking between semantic arguments and their
position in the syntax has two major phases:
The core arguments are those that are determined by the semantic decom-
position of the verb. Adjuncts are non-argument PPs and adverbs, which
occur in a periphery. RRG recognizes a third class of argument, the ‘argu-
ment-adjuncts’, which will be presented in 6.4.
The syntactically relevant semantics of a sentence is represented in the LS
(see Chapter 3 of this volume). That semantic representation determines the
kind of syntactic template (e.g. tree structure) that the sentence maps onto,
following a default principle.
The number of syntactic slots for arguments and argument-adjuncts within the
core is equal to the number of distinct specified argument positions in the
semantic representation of the core. (Van Valin 2007: 130)
The standard mapping of a verb’s arguments onto the actor and undergoer
macroroles follow the Actor-Undergoer Hierarchy (AUH; Chapter 4). There
are, however, processes that are common cross-linguistically that alter the
standard argument structure or alignment. These include the two types of
constructions discussed in this section: noun incorporation and the ditran-
sitives or dative constructions.
In (4a) the verb is clearly transitive, requiring the direct object pronom-
inal form. The verb in (4b), however, is intransitive and so lacks the
direct object marking. Now consider the following examples from
Tongan (Polynesian) (5a–b) and Yucatec Maya (6a–b) (from Mithun 1984:
851, 857).
The person affected can assume ‘a primary case role’ in these constructions
because the transitivity of the verb is unchanged, with the possessor of the
body part linking to the undergoer macrorole, as the ʻaffected personʼ.2
Mithun (1984: 857) provides the following example from Tupinamba (Tupí).
In both examples, the first person singular prefix a- maps onto the actor
macrorole. In (8a) ‘his face’, as patient, is the unmarked choice for under-
goer. The incorporated form in (8b) requires the marked choice for under-
goer and the possessor maps onto the undergoer macrorole.
One other kind of noun incorporation needs to be briefly discussed here.
At the beginning of this section, it was mentioned that Greenlandic Eskimo
is known to have extremely productive noun incorporation (Sadock 1991).
A language with similar constructions is Southern Tiwa (Kiowa-Tanoan) and
both languages are discussed in Rosen (1989). Consider the following
examples from Allen et al. (1984: 297).
These examples, like many in Allen et al. (1984) and Sadock (1991), are
notably different from the noun-incorporation examples discussed previ-
ously, in at least two ways. First, the incorporated nouns are modified by a
quantifier (9a) and a deictic (9b), and, second, they are clearly referential.
Van Valin and LaPolla (1997: 66–68) present an analysis of similar construc-
tions in Greenlandic Eskimo, showing how RRG accommodates such struc-
tures with distinct operator and constituent projections in the layered
structure of the noun phrase. The significant conclusion is that, unlike the
earlier examples discussed in this section, these do not involve the incorpor-
ation of a bare noun, but rather full RPs, as shown by the fact that they are
referential and by the presence of modifiers.
6.3.2 Ditransitives
For our purposes here, I will assume the definition of ditransitives offered by
Malchukov et al. (2010: 1):
Following the AUH, the first argument of do′, x, is the actor, while the
theme, the second argument of have′, z, is the undergoer.
Over the last fifty years, there have been many studies of the English
dative alternation that try to account for the semantic similarities and
differences between pairs of sentences such as those in (2) and (11).
The verb agrees with the actor and the undergoer in the simple transitive
clauses (16a) and (16b) and with the recipient in the ditransitive clause (16c).
This agreement pattern in the ditransitive is obligatory – the alternation
found in English dative movement is absent. (For further discussion of
Tibeto-Burman examples, see Van Valin and LaPolla 1997: 272–273.) In
‘primary object languages’, unlike the Mixtec and French examples, the
recipient is obligatorily linked to the undergoer macrorole.
In Van Valin and LaPolla (1997: 387) this marked choice for undergoer is
attributed to animacy – the animate indirect object is chosen over the
inanimate direct object for undergoer position (see, however, Guerrero
and Van Valin 2004 for further discussion).
Some languages, such as Mixtec and Tepehua, have no syntactically
ditransitive verb roots. Although Tepehua has no verb roots with three
direct core arguments, the LS for predicates such as ma:laqatʃa: ‘send’ (17a)
and stʼa: ‘sell’ (18), like their counterparts in other languages, require three
semantic arguments. Consider the examples in (17) and (18).
(17) a. ma:laqatʃa:-ɬ
send-pfv
‘s/he sent it’
b. ma:laqatʃa:-ni-ɬ
send-dat-pfv
‘s/he sent it to him/her’
(18) a. ʃtaq-ɬi
give-pfv
‘s/he gave it.’
b. ʃtaq-ni-ɬ
give-dat-pfv
‘s/he gave it to him/her’
(19) a. [do′ (x, Ø)] CAUSE [BECOME have′ (Ø, z)] (¼ (17a), (18a))
b. [do′ (x, Ø)] CAUSE [BECOME have′ (y, z)] (¼ (17b), (18b))
The translation of (20b) shows that the antipassive suffix -nVn in Tlachichilco
Tepehua results in deletion of the patient undergoer. In (20c), the presence
of the applicative suffix -ni, marks the recipient as undergoer. This is evi-
denced by the fact that it is the recipient that is deleted in (20d), when the
antipassive suffix is present.
In sum, RRG posits only two macroroles even though predicates may have
more than two arguments in their logical structure. Ditransitive construc-
tions display different approaches to the syntactic position of the recipient
argument. In some languages, the theme is the undergoer, following the
default choice on the AUH. In other languages, the recipient is regularly the
marked choice for undergoer, reflecting its prominence as the typically
more animate non-actor argument. Finally, other languages allow variable
linking to the undergoer macrorole.
It has often been noted that this alternation reflects a semantic distinction
in that the NP or RP immediately following the verb shows a higher level of
affectedness: as Fillmore notes (1968: 48, fn.49), the sentence in (22b) implies
the entire wall got painted but (22a) does not. Furthermore, in the corres-
ponding passive sentences, the semantic distinction is maintained, showing
that the difference in affectedness is due to undergoer status, not to the
syntactic position of direct object.
6.3.4 Causative
Lexical causative constructions built on intransitive verbs are generally
straightforward when it comes to argument structure: the causee is linked
to undergoer in both the intransitive and transitive constructions; in the
transitive construction, the causer links to actor.
(27)
transitive causative of transitive
ʔah-ya ‘s/he digs it’ maːʔah-ni:-y ‘s/he makes her/him/it dig it’
ʃʔoq-ya ‘s/he unties it’ ma:ʃʔoq-niː-y ‘s/he makes him/her/it untie it’
In clear contrast to the pattern in the West Coast Bajau and Tepehua
examples is causative formation on a transitive verb in which the undergoer
of the base verb retains the syntactic marking and behaviour of undergoer.
The causee then appears as a peripheral or indirect core argument. An
example of this can be seen in the Turkish sentences in (28) (from Underhill
1976: 346):
In both the basic transitive as well as the causative construction, the direct
object, dis, ‘tooth’ is marked as accusative and, in the causative, the causee
˙
must be marked as dative.
Like other lexical constructions in this section, morphological causatives
affect the level of mapping from predicate argument structure onto the
actor and undergoer macroroles. For causatives formed on a transitive verb
base, two non-actor core arguments are present in the LS, and languages
differ in regard to which of the two links to the undergoer macrorole.
6.3.5 Anticausative
The anticausative construction has been described as the inverse of the
causative: instead of adding a causer and causing event to the LS, the antic-
ausative is a derived intransitive in which the undergoer is the PSA. Haspel-
math and Müller-Bardey (2001) call the anticausative ‘the most radical
agent-removing category’ and include examples such as the following from
Hungarian and Turkish.
In each of these examples, the verb root is causative with actor and under-
goer and the intransitive is derived by a morphological operation.
The RRG literature discusses two kinds of anticausatives. The alternations
in (29) and (30) display an anticausative that is similar to the ‘middle’
construction in traditional grammar. In these constructions, ‘the function
of the morphological markers is to cancel part of the logical structure’ (Van
Valin 2005: 46) – the anticausative removes the causer and causing event
from an otherwise causative verb. Typically, the anticausative alternation
applies to a minor subclass of verbs. This is the case in Tepehua languages,
which have a small group of transitive verb roots that have derived intransi-
tive forms marked by the inchoative prefix, ta-, as in the following examples.
(31)
Base verb Gloss Derived form Gloss
(causative) (anticausative)
The obvious similarity between the passive and this anticausative con-
struction is that both remove the actor as a core argument. However, there is
a key difference between them: the passive has an implicit or understood
actor and causing event but this anticausative construction does not. In RRG
this means that, unlike the passive, in which the actor is implicit but not a
core argument, in these anticausatives the actor and causing event are
missing altogether.
González Vergara shows that (33b), like the Italian examples in Centineo
(1996), has an implicit causing agent and can occur with manner adverbs.
Thus, unlike the anticausatives accounted for by (32), these maintain the
causing activity in the LS.
Both of the rules in (32) and (34) change an event with two specified
arguments into an achievement or accomplishment with one undergoer
but they have significantly different logical structures.
To account for this common alternation between atelic and telic uses of
some verbs, ‘not just verbs but in fact whole verb phrases must be taken
into account to distinguish activities from accomplishments’ (Dowty
1979: 60–62). The RRG analysis of the examples in (35) and (36) requires
distinct logical structures for the atelic and telic readings of the verbs.
The objects in (35a) and (36a) are non-referring NPs but those in (35b) and
(36b) are RPs. As a result, there is a difference in macrorole status, since
activities, including multiple-argument activity verbs (the atelic
examples), never have an undergoer macrorole. A common explanation
for the data in (35) and (36), then, is that the alternation is due to the
change in the referential status of the object, that is, whether it is an RP
or NP.
Van Valin (2012: 69–71) shows that the activity–active accomplishment
alternation in other languages often is not due to the inferred status of the
object, but is marked on the verb. Changing the aspect marking on the verb
can result in a change from activity to active accomplishment (‘He was
eating everything for two hours,’ vs. ‘He ate everything in two hours.’) The
following lexical rules account for the alternations, showing the change in
the LS (see also Van Valin and LaPolla 1997: 180 and, for a revised LS of active
accomplishments which has no consequences for the current discussion,
Van Valin 2018 and Bentley 2019).
These lexical rules reflect the productivity of an alternation between two LSs
involving the same base verb.
To show that this alternation cannot be attributed simply to the presence of
a definite direct object, Van Valin (2012: 70) presents examples from Georgian
(Holisky 1981), in which the preverb da- imposes a telic reading of the event.
The kind of alternation accounted for in the rule in (37) is not marked
morphologically in English but is in languages such as Georgian and it
is marked by aspectual markers in other languages such as Russian. The
distinct logical structures of activities and their corresponding active
accomplishments have both semantic and syntactic consequences.
6.4.1 Benefactive
Though often treated as a subtype of the dative construction, a benefactive
typically has a different relation to the event referred to by the verb. In the
case of ditransitives, that is, constructions in which the third argument in
some sense ‘receives’ the second one, the verb itself introduces an LS that
includes the three arguments. This is not the case with a benefactive
construction. A benefactive construction involves an argument external
to the event. Unlike the semantic representation of an adjunct, the seman-
tic representation of a benefactive shares an argument with the LS of
the verb.
In RRG, the benefactive involves purpose, and can be represented by the
semantic representation given for ‘purposive for’ in English in Van Valin and
LaPolla (1997: 383) (based on Jolly 1991).
(41) [want′ (x, LS2 ^ DO (x,[make′ (x, house)]) CAUSE [have′ (I, house)]
(43) ʔik-miɬpaː-ni-ya-n
1sbj-sing-dat-fut-2obj
‘I will sing for you.’
Baker points out that in this construction, the applicative -ir has resulted
in the benefactive argument usurping properties associated with the direct
object: immediate postverbal position, object pro-drop, and (in this
example) subject of the passive (compare 44a and 44b). In RRG terms, this
ˇ
is evidence that in the Chichewa applicative construction, the benefactive
is the undergoer. As we saw in 6.4.1, this is not surprising, as the benefac-
tive is typically an example of an argument-adjunct rather than a
simple adjunct.
However, some applicatives allow adjuncts to appear as syntactic argu-
ments of the verb. As reported in Peterson (2007: 18, 19, 22), Hakha Lai, a
Tibeto-Burman language, has several optional applicative constructions that
allow an adjunct to appear as a verbal argument.
(47) a. kin-ta-tʼaː-ʔa-ɬ
1obj-3pl.sbj-com-go-pfv
‘They went with me.’
b. puː-mi-ɬ huːki
ins-come-pfv horse
‘S/he came by horse.’
c. waː yuːtʃa ɬiː-stʼaː-ɬ
foc 3pron dir-sell-pfv
‘S/he sold it for that (price) / sold it for that (reason).’
(48) a. Does the argument of the base verb or the argument of the applicative link
to the undergoer macrorole?
b. Is there a restriction regarding which of the non-actor arguments may
occur as PSA in a passive construction?
applicatives, ɬiː-, puː-, tʼaː-, most often simply allow adjuncts to appear as
syntactic arguments of the verb – an important feature since only direct
arguments of the verb can be questioned or relativized in Tepehua.
However, the linking of the PSA in a passive construction is not limited
to the undergoer: any direct syntactic argument of the verb can be
the PSA.
6.5 Voice
6.5.1 Passive
According to Keenan and Dryer (2007: 328–329), the following passives are
‘basic passives’ ((51) is their example, (52) is Tlachichilco Tepehua).
This is considered to be a ‘basic’ passive because passives like this ‘are the
most widespread across the world’s languages’. They have the following
characteristics: (i) there is no agent phrase (in Tepehua, as in many other
languages, the passive does not allow the actor to appear in the clause at
all); (ii) the verb is a transitive verb that is passivized; (iii) in its non-
passivized form, the verb expresses an action with an agent subject and
patient object.
By these criteria, then, English passives with an explicit actor as well as
impersonal passives (passives of intransitives) are not basic passives. The
contrast between the two Tepehua forms in (53) illustrates the use of the
passive with intransitive verbs, as do the examples from German and
Turkish in (54a) and (54b), respectively.
(55) a. k-laqtsʼin-kan-a:-w
1sbj-see-pass-ipfv-1pl
‘We are seen.’
b. kin-ta-laqts’in-kan-a:-n
1obj-3plsbj-see-pass-ipfv-2obj
‘We are seen.’
In the passive construction in (56a) the undergoer assumes PSA status; in the
passive in (56b), however, the woman is marked by the directional postpos-
ition, -u, and is not linked to the undergoer position. As a result, the PSA
modulation of the passive does not apply, only the argument modulation,
removing the actor.
However, some languages allow a wide variety of arguments to link to
the PSA position. Van Valin (2005: 121) gives examples from Kinyarwanda
(Bantu), in which applicatives can mark several semantic roles. Tepehua is
another example. The Totonac-Tepehua languages have pragmatically
determined word order and no case marking on the NPs. It is not possible
to determine, for a transitive clause in isolation with two third-person
singular participants, which is the PSA. The passive and antipassive play
a major role in tracking referents. In (57), the instrumental and comitative
appear as syntactic arguments of the verb due to the applicatives though
they do not have macrorole status. However, in the passive they can map
onto the PSA.
6.5.2 Antipassive
The term ‘antipassive’ refers to a change in syntactic alignment of argu-
ments that has traditionally been used to describe a detransitivizing form in
ergative languages, such as in the following examples from Dyirbal (from
Dixon 1994: 161, 170).
Considering the two aspects of voice in RRG (49), the same can be said for
antipassives as for passives: in some languages the relevant construction
only affects the argument modulation and does not directly affect the PSA
modulation. In fact, many ergative languages are only ergative in their
morphology and do not have the syntactic ergativity found in a language
like Dyirbal.
In a non-ergative language like Tepehua, the antipassive only involves
the level of argument modulation (49b). It occurs on a transitive verb base,
marking the absence of the undergoer, whether it is the undergoer of the
verb root or the marked undergoer, that is, the argument of the dative
suffix -ni. But it never marks the absence of the applicative argument
associated with one of the applicative prefixes (comitative, instrumental,
directional).
6.6 Conclusion
References
Allen, Barbara J., Donna B. Gardiner and Donald G. Frantz. 1984. Noun
incorporation in Southern Tiwa. Indonesian Journal of Applied Linguistics
50: 292–311.
Levin, Beth. 1993. English Verb Classes and Alternations: A Preliminary Investi-
gation. Chicago: University of Chicago Press.
Malchukov, Andre, Martin Haspelmath and Bernard Comrie. 2010. Studies
in ditransitive constructions: A typological overview. In Andre Malchukov,
Martin Haspelmath and Bernard Comrie (eds.), Studies in Ditransitive Con-
structions. A Comparative Handbook, 1–64. Berlin: De Gruyter, Mouton.
McGregor, William. 1997. Grammatical structures in noun incorporation. In
Anne-Marie Simon-Vandenbergen, Kristin Davidse and Dirk Noël (eds.),
Reconnecting Language: Morphology and Syntax in Functional Perspectives. Cur-
rent Issues in Linguistic Theory, 141–180. Amsterdam: John Benjamins.
Miller, Mark. 2007. A Grammar of West Coast Bajau. PhD dissertation, (Arling-
ton), University of Texas.
Mithun, Marianne. 1984. The evolution of noun incorporation. Language 60:
847–894.
Perlmutter, David M. and Paul M. Postal. 1977. Toward a universal charac-
terization of passive. In K. Whistler, R. D. Van Valin et al. (eds.), Proceedings
of the Annual Meeting of the Berkeley Linguistics Society, 3, 394–417. Berkeley,
CA: Berkeley Linguistics Society.
Peterson, David A. 2007. Applicative Constructions. Oxford: Oxford University
Press.
Polinsky, Maria. 2013. Antipassive constructions. In Matthew S. Dryer and
Martin Haspelmath (eds.), The World Atlas of Language Structures Online.
Leipzig: Max Planck Institute for Evolutionary Anthropology. http://wals
.info/chapter/108.
Rosen, Sarah Thomas. 1989. Two types of noun incorporation: A lexical
analysis. Language 65: 294–317.
Sadock, Jerrold M. 1991. Autolexical Syntax: A Theory of Parallel Grammatical
Representations. Chicago: University of Chicago Press.
Sapir, Edward. 1911. The problem of noun incorporation in American lan-
guages. Language 13(2): 250–282.
Underhill, Robert. 1976. Turkish Grammar. Cambridge: MIT Press.
Van Valin, Robert D. Jr. 2005. Exploring the Syntax–Semantics Interface. Cam-
bridge: Cambridge University Press.
Van Valin, Robert D. Jr. 2007. The Role and Reference Grammar analysis of
three-place predicates. Suvremena Lingvistika 33.1(63): 31–64. https://rrg.caset
.buffalo.edu/rrg/vanvalin_papers/RRG-Analysis_Three-Place_Pred.pdf.
Van Valin, Robert D. Jr. 2012. Lexical representation, co-composition, and
linking syntax and semantics. In James Pustejovsky, Pierrette Bouillon,
Hitoshi Isahara, Kyoko Kanzaki and Chungmin Lee (eds.), Advances in
Generative Lexicon Theory: Text, Speech and Language Technology, Vol. 46,
67–107. Dordrecht: Springer.
Van Valin, Robert D. Jr. 2018. Some issues regarding (active) accomplish-
ments. In Rolf Kailuweit, Lisann Künkel and Eva Staudinger (eds.), Applying
and Expanding Role and Reference Grammar, 71–93. Freiburg: Freiburg Insti-
tute for Advanced Studies, Albert-Ludwigs Universität Freiburg.
Van Valin, Robert D. Jr and Randy LaPolla. 1997. Syntax: Structure, Meaning,
and Function. Cambridge: Cambridge University Press.
Vendler, Zeno. 1967[1957]. Linguistics in Philosophy. Ithaca, NY, Cornell Univer-
sity Press. (Previously published in Vendler, Zeno. 1957. Verbs and times.
Philosophical Review 56: 143–160.)
Watters, James K. 2017. Tlachichilco Tepehua: Semantics and function of
verb valency change. In Álvaro González and Ía Navarro (eds.), Verb Valency
Changes: Theoretical and Typological Perspectives, Typological Studies in Lan-
guage 120, 165–192. Amsterdam: John Benjamins.
Watters, James K. 2019. La preposición en tepehua y construcciones seme-
jantes. In Lilián Guerrero (ed.), Adposiciones y elementos de su tipo en lenguas
de América. Estudios sobre Lenguas Americanas 9, 315–344. Mexico: Uni-
versidad Nacional Autónoma de México, Instituto de Investigaciones
Filológicas.
Williams, Alexander. 2015. Arguments in Syntax and Semantics. Cambridge:
Cambridge University Press.
Zylstra, Carol F. 1991. A syntactic sketch of Alacatlazala Mixtec. In Bradley, C.
Henry and Barbara E. Hollenbach (eds.), Studies in the Syntax of Mixtecan
Languages, Vol. 3, 1–178. Dallas: Summer Institute of Linguistics and the
University of Texas at Arlington.
Notes
1 When citing research done by others, I have tried to keep the same
morpheme glosses that they have used.
2 For discussion of body-part incorporation with an intransitive verb,
resulting in a change of undergoer assignment (‘possessor raising’), see
Van Valin (2005: 145–146).
3 The gloss B refers to one of three gender classes, in this case, one that
refers to either an animate plural or an inanimate singular (Allen et al.
1984: 293, fn. 5)
4 The passive or ‘unspecified subject’ construction in Tepehua does not
provide evidence of undergoer status. In a Tepehua passive construction,
a non-undergoer can map onto the PSA (see 6.4.2).
5 These parallel the ‘two universals of passivization’ presented by Perlmut-
ter and Postal: (i) ‘A direct object of an active clause is the (superficial)
subject of the “corresponding” passive,’ and (ii ‘The subject of an active
clause is neither the (superficial) subject nor the (superficial) direct object
of the “corresponding” passive’ (1977: 399).
Abbreviations
7.1 Introduction
The aim of this chapter is twofold: to explicate the theory of case assignment
in Role and Reference Grammar (RRG; Van Valin 1991, 1993, 2005, 2009; Van
Valin and LaPolla 1997) and to outline its Optimality-Theoretic (OT) imple-
mentation (Nakamura 1999a, 1999b) and its extension to instrumental case
assignment and case syncretism (Nakamura 2011, 2021).
Case in RRG constitutes part of the linking system in which verbal argu-
ments are realized by case/cross-referencing markers, agreement markers,
CLAUSE
CORE PERIPHERY
RP NUC RP MP
PRED
V ADV
projection and the operator projection. The former projection represents the
clause structure with three nested layers, the nucleus (the predicate), the
core (the predicate and its arguments), and the clause (the core and any
peripheral elements that modify the core) (as illustrated in Figure 7.1), while
the latter consists of auxiliary elements (e.g. aspect, tense, root/epistemic
modal, status, evidential, speech act) that are hierarchically and topologic-
ally ordered according to the layer they modify.
(2) a. State
The cup is shattered. shattered′ (cup)
Carl is in the library. be-in′ (library, Carl)
b. Activity
The children cried. do′ (children, [cry′ (children)])
Kim ate fish. do′ (Kim, [eat′ (Kim, fish)])
c. Achievement
The windows shattered. INGR shattered′ (windows)
The balloon popped. INGR popped′ (balloon)
d. Semelfactive
Dana glimpsed the dog. SEML see′ (Dana, dog)
Mary coughed. SEML do′ (Mary, [cough′ (Mary)])
e. Accomplishment
The snow melted. BECOME melted′ (snow)
Chris learned French. BECOME know′ (Chris, French)
f. Active accomplishment
John ran to the park. do′ (John, [run′ (John)]) & INGR be-at′ (park, John)
Kim ate the fish. do′ (Kim, [eat′ (Kim, fish)]) & INGR consumed′ (fish)
g. Causative
The dog scared the boy. [do′ (dog, ø)] CAUSE [feel′ (boy, [afraid′])]
Max melted the ice. [do′ (Max, ø)] CAUSE [BECOME melted′ (ice)]
Felix bounced the ball. [do′ (Felix, ø)] CAUSE [do′ (ball, [bounce′ (ball)])]
(3a) determines the number of macroroles a verb takes, while (3b) deter-
mines which macrorole (actor or undergoer) it is when the verb receives only
one macrorole. When the number of macroroles does not follow from the
DMAP, it has to be specified in the lexical entry of the verb by a feature [MRα]
(where α represents the number of macroroles).
The relationship between LS argument slots (or thematic relations serving
as mnemonics for them) and macroroles is captured by the Actor-Undergoer
Hierarchy (AUH) in Figure 7.2.
Actor Undergoer
The AUH states that given the LS of a multi-argument verb, the leftmost
argument will be the actor and the rightmost argument will be the
undergoer.
(4a)–(4f) illustrate the default macrorole assignment.
The observation that the locative argument in (5b) (the truck) is construed as
being fully loaded motivates its choice as undergoer.2
Finally, the privileged syntactic argument (PSA) selection is based on the
PSA Selection Hierarchy in (6), the PSA selection principles in (7), and
restrictions in (8) (Van Valin 2005: 100).
The following shows how (9) and (10), respectively, work in Icelandic and
Warlpiri and how the macrorole-based account of case assignment extends to
active(-stative) case systems. First, let us consider the Icelandic data given in (12).
(13) Icelandic (Andrews 1990: 210; Van Valin 1991: 174, 175)
a. Stráknum líkar slíkir bílar.
the.boy.dat likes such cars.nom
‘The boy likes such cars.’
b. Henni hefur alltaf þótt Ólafur leiðinlegur.
her.dat has always thought Olaf.nom boring.nom
‘She has always considered Olaf boring.’
c. Ég skilaði henni peningunum.
I.nom returned her.dat the.money.dat
‘I returned her the money.’
The DMAP predicts that all of the multiple-argument verbs in (13a)–(13c) take
a pair of actor and undergoer. However, this prediction is not borne out;
contrary to (3a1), all of (13a)–(13c) contain only one macrorole argument.
The case assignment rules in (9) apply to the combination of the non-
macrorole and undergoer arguments and yield the DAT-NOM case frame
in (13a). The same account holds for the DAT-NOM case frame in (13b).
Likewise, suppose that skila ‘return’ in (13c) is lexically specified for having
only one macrorole ([MR1]). (3b) requires that the only macrorole that the
verb takes is an actor, since it has an activity predicate do′ in its LS in (15).
LS: [do′ (I, ø)] CAUSE [INGR have′ (she, money) ^ NOT have′ (I, money)]
This macrorole assignment accounts for why the non-PSA arguments in (13c)
receive dative case. Lexical specification of the number of macroroles that
þykja ‘think’, líka ‘like’, and skila ‘return’ involve allows RRG to treat dative
not as an example of quirky cases, but as one of the regular cases. This is one
of the major points of contrast between RRG and the other syntactic theories
with respect to case assignment.
(9) also accounts for the contrast between (17) and (18) with respect to case
preservation under passivization.
What is intriguing about (19) is that the theme argument of þykja ‘think’ in
the dependent core receives nominative case. This Icelandic example casts
doubt on the assumption shared by major syntactic theories that nomina-
tive case assignment may occur only in finite clauses and suggests the
necessity of making it an option for a language to allow (9) or (10) to apply
in each core within a clause independently.
Van Valin and LaPolla (1997: 575–581) propose to derive the nominative case
assignment in the dependent core in (19) from the DCA parameterization. The
fact that the theme argument in the dependent core receives nominative case
suggests that the DCA for Icelandic is the core, in contrast to languages such as
German, which allows no nominative-marked argument to occur in any
dependent core.7 (20) is a summary of the above discussion.
The assumption that the DCA for Icelandic is the core explains the DAT-
NOM case frame of the dependent verb in (19). First, let us assume that the
dependent verb in (19) is irregular with respect to macrorole transitivity
and that the two-place verb þykja ‘think’ has the feature [MR1] in its lexical
entry. (3b) requires the only macrorole to be an undergoer, since þykja
‘think’ has no activity predicate in its LS. The AUH requires the theme
argument to be an undergoer, which leads the experiencer argument of
þykja to become a non-macrorole core argument. On the other hand, the
experiencer argument of the matrix verb telja ‘believe’ is an actor and
counts as ‘the highest-ranking macrorole argument’ in (9a). (21) shows
how the macrorole assignment proceeds in the matrix and dependent core
of (19).
Applying the set of case assignment rules in (9) to the dependent core in (21)
accounts for the nominative marking of the theme argument (Ólafur) in (19).
In contrast to (9), which derives accusative case systems, (10) accounts for
ergative case systems. (22a)–(22c) come from Warlpiri (Pama-Nyungan).
b. Geu¼jak (gopnyan)
3¼go (3sg)
‘(S)he goes.’ (Actor)
c. Lôn ehët(¼lôn)
1sg fall(¼1sg)
‘I fall.’ (Undergoer)
Two points are worth making about (27). First, the implement with is distinct
from the instrument with, in that the instrument argument is part of a
verb’s causal chain, while the implement argument is not. Second, with
exhibits an instrumental/comitative syncretism.
Jolly (1991, 1993) makes an important observation that the instrument
with marks a potential actor that fails to function as actor, while the
locatum with marks a potential undergoer that fails to serve as undergoer
and that the comitative with marks one of the co-participants that would
otherwise appear as an actor or undergoer but does not. (28) is the unified
definition of the instrument, locatum and comitative uses of with.11
The basic idea is that all of the instrument, locatum and comitative uses
of with are non-predicative: they are associated with the outcome of certain
linking options (as described in (28)), in contrast to locative prepositions (as
illustrated in (29)), which are linked to a specific LS configuration.
The two prepositions in (29) are associated with their LSs: in introduces the
LS ‘be-in′ (. . .)’, a two-place predicate that takes the entire event (‘[do′ (John,
[run′ (John)])]’) as the second argument and situates it in a location desig-
nated by the first argument, while to introduces the LS ‘INGR be-at′ (. . .)’,
whose first argument designates the goal of the movement.
Three remarks are in order about the ICR in (28). First, (28) does not extend
to the implement with illustrated in (27b). The fact that the two-place activity
verb write has no ready-made argument slot for the implement NP a pen
disqualifies it as a possible candidate for an actor and puts the implement
use of with outside the scope of (28). Second, (28) interacts with the dative case
assignment rule in (9c) in an intricate way. For example, (27c) (repeated
below) involves a marked undergoer assignment to the locative argument,
which forces the theme argument hay to become a non-macrorole.12
The problem is that both (9c) and (28) may apply to hay. Van Valin (2005)
argues that the ICR in (28) overrides (9c) and accounts for why hay is marked
by the instrumental preposition under the crucial assumption that both
instrumental with-phrases (as illustrated in (27a)) and locatum with-phrases
(as illustrated in (27c)) constitute a subset of non-macrorole core arguments
and that (28) applies to them.13 Finally, (28) applies to nominal cases and
adpositions that nullify the morphological distinction between instrumen-
tal and comitative, but it is important to note that their syncretism is not
common outside the Indo-European languages (Stolz et al. 2006). The fact
that the majority of languages outside the Indo-European family distinguish
the two cases (as illustrated in (30)) suggests an alternative strategy to
analyse the non-comitative and comitative uses of with separately and then
(30) Japanese
a. John-ga katana-de take-o kit-ta.
John-nom sword-ins bamboo-acc cut-pst
‘John cut the bamboo with a sword.’ (Instrumental)
b. John-ga Tom-to/*de eiga-o mi-ta.
John-nom Tom-COM/ins movie-acc watch-pst
‘John watched the movie with Tom.’ (Comitative)
c. John-ga kome-o niku-to/*de maze-ta.
John-nom rice-acc meat-com/ins mix-pst
‘John mixed rice with meat.’ (Comitative)
7.4 An OT Implementation
Let us see how the constraint ranking in (33a) derives the NOM-ACC and
DAT-NOM case frames in Icelandic. Tableau 7.1 shows how (33a) outputs the
NOM-ACC case frame illustrated in (12b).17
The first and second candidates violate the top-ranking constraint (31d),
since the non-macrorole core argument receives nominative case. The fourth
candidate violates (31a), since it contains no nominative argument. The
above consideration leaves the third candidate as the winner. The third
candidate violates (31b), but it fares better than the other candidates, since
the other candidates violate either (31d) or (31a) (which is ranked higher
than (31b)).
An analogous account holds for ergative and active(-stative) case systems.
Tableau 7.3 shows how the ERG-NOM case frame is derived in (22c).
The difference between Tableau 7.1 and Tableau 7.3 comes down to the relative
ranking of (31b) and (31c): when (31b) outranks (31c), an accusative case system
emerges, while when (31c) outranks (31b), an ergative case system emerges.
This suggests that the traditional distinction between nominative and absolu-
tive is a by-product of the relative ranking of (31b) and (31c) and that there is no
need to postulate absolutive case in addition to nominative case.
Finally, the constraint rankings in (33d) and (33e) yield accusative-active
and ergative-active case systems, respectively. When (31b) outranks (31a) as
in (33d), undergoer arguments always receive accusative marking, while
actor arguments receive nominative marking. In contrast, when (31c) out-
ranks (31a) as in (33e), actor arguments receive ergative case, while under-
goer arguments receive nominative case.
To summarize this subsection, the OT reformulation of the case assignment
rules in (9), (10), (25) and (26) in terms of the case assignment constraints in
(31) accommodates the major case frames of simple clauses in accusative,
ergative, and two types of active case systems and derives the typological
variation of case systems from re-ranking of the individually simple con-
straints that make no reference to the relative ranking of actor and undergoer.
Nominative
Accusative þ
Dative þ þ
Instrumental þ
(34) Russian (Kilby 1977: 75; Wierzbicka 1980: 23; Janda 1993: 147, 156)
a. Ivan napisal pisʹmo ručkoj.
Ivan.nom wrote letter.acc pen.ins
‘Ivan wrote the letter with a pen.’
b. Ivan pil vino litrami.
Ivan.nom drank wine.acc liters.ins
‘Ivan drank wine by the litre.’
c. Petr kivnul golovoj.
Peter.nom nodded head.ins
‘Peter nodded his head.’ (lit. ‘Peter nodded with the head.’)
d. Monax dolžen svjazatʹ usta svoi molčaniem.
monk.nom must tie.up lips.acc own.acc silence.ins
‘A monk must seal his lips with silence.’
e. Lošadʹ soseda lučše moej
horse.nom neighbour.gen better mine.gen
i siloj i krasotoj
and strength.ins and beauty.ins
‘The neighbour’s horse is better than mine both in strength and in appearance.’
f. Oni gruzili baržu drovami.
they.nom loaded barge.acc firewood.ins
‘They loaded the barge with firewood.’
In order to treat instrumental case on a par with the regular cases (i.e.
nominative, dative, accusative and ergative), it is imperative to derive all
the argument and adjunct uses of instrumental case from a single con-
straint associated with a unitary meaning. Second, the question of how
the instrumental case assignment interacts with the dative case assignment
needs to be addressed. What is at stake here is how to explain why hay in
(34f) receives instrumental marking despite being qualified to receive dative
case because of its status as a non-macrorole core argument. This is a clear
indication that it is not enough to define instrumental as the default case
for adjuncts. Finally, it remains to be shown how to derive the wide range of
interpretations an instrumental-marked noun may have from its
unitary meaning.
In order to solve the first problem, Nakamura (2015, 2021) defines instru-
mental case as the default case for everything other than macrorole argu-
ments and proposes (35b) as a constraint that represents the unitary
Invoking the QS of pen (more specifically, its telic quale) for interpreting the
prepositional phrase (and the whole sentence that contains it) in (27e) leads
to introducing a new argument position for its complement noun (pen) and
thereby prevents a violation of the CC. Van Valin (2012) derives the LS of (27e)
(given in (37c)) by merging the LS in the telic quale in (37a) with the LS of the
active accomplishment use of the activity verb write in (37b). This operation
is termed co-composition (Pustejovsky 1995), an operation that derives the
meaning of a phrase compositionally from the head (in this case, the
preposition) and its argument when the latter affects the meaning of the
phrase (and the clause that contains it) beyond its role as an argument of the
head. An analogous account holds for (34a).
Second, Nakamura (2021) takes the above QS-based account of (27e) as the
first step toward identifying contextually appropriate interpretations of
instrumental-marked nouns and extends it to other uses of instrumental-
marked nouns, four of which are given in (34b)–(34e). Let us begin with (34b),
which involves a measure unit by which to quantify an indefinite amount of
some object, while engaging in some activity involving it (in this case,
drinking wine). We may take the instrumental-marked noun litrami ‘litres’
as an adjunct, since it is not subcategorized by the verb. The underspecified
meaning of instrumental case requires the unit noun litr ‘litre’ to be linked
to an argument position of an event involved by its QS. I propose that the
unit interpretation of litrami is derived from its telic qualia given in (38a): litr
‘litre’ refers to a unit that is used to measure an amount of liquid.
Peter’s action (Wierzbicka 1980). Second, the body-part noun forms a posses-
sive relation with the subject argument: its constitutive quale (‘have.as.part′
(x:human, y)’) allows the hearer to associate Petr and golovoj with the posses-
sor (‘x’) and possessum (‘y’), respectively. Another point to note in this
connection is that (34c) encodes the possessive relation in terms of predica-
tion (and not in terms of reference, as illustrated in (39)) despite the fact that
the possessive predicate remains covert.
The main predicate kivnutʹ ‘nod’ and the covert possessive predicate share
the subject argument Petr, but these predicates do not form a nuclear
juncture, since Petr participates in the two events in (40a,b) independently.22
This suggests that (34c) embodies a non-subordinate core juncture, in which
each of the predicates has its own core.
The macrorole assignment proceeds in the two cores independently. The
main predicate kivnutʹ ‘nod’ is a single-macrorole verb. Petr functions as
undergoer when it counts as the first argument of the covert possessive
predicate in (40b) (since it has no activity predicate do′ in its LS), while Petr
serves as actor when it counts as the effector argument of the main predi-
cate. The possessum argument golova ‘head’ becomes a non-macrorole core
argument and receives instrumental case from (35b), since the possessor
argument Petr bears the undergoer status in violation of the AUH.23
In contrast to (34a)–(34c), the instrumental-marked noun in (34d) (molčanie
‘silence’) has no (default) value for its constitutive, telic or agentive quale: it
is not conventionally associated with any particular material, purpose,
function or origin. This yields the consequence that we need to understand
the function of molčanie with no reference to its QS. Since molčanie is neither
an argument of the predicate nor part of the predicate, we have to conclude
by a process of elimination that this instrumental-marked noun serves as a
modifier: it modifies the manner of the action denoted by the verb svjazatʹ
‘tie up.’
Finally, let us consider how the meaning of domain restriction is derived
in (34e). The first point to note here is that both sila ‘strength’ and krasota
‘beauty, appearance’ are part of the formal qualia of lošadʹ ‘horse.’ The fact
that these attributes constitute part of the attributes of a horse explains why
the two instrumental-marked abstract nouns serve to relativize the evalu-
ation of the neighbour’s horse by the speaker and receive the interpretation
of domain restriction. The fact that the two abstract nouns are not subcat-
egorized by lučše suggests that they function as a modifier of the predicate
and receive instrumental case from (35b) as dominated by (35a) (since they
are not syntactic arguments).
One of the major principles of the RRG theory of case assignment (Van Valin
1993, 2005; Van Valin and LaPolla 1997) is that it does not distinguish syntactic
cases from morphological cases, under the assumption that each case receives
its unique morphological realization (see Section 7.1). However, this assump-
tion is called into question by case syncretism phenomena, where a single case
morpheme realizes more than one syntactic case. This section outlines the OT-
RRG account of case syncretism (based mainly on Nakamura (2011)), with a
particular focus on those instances of syncretism in which a single case
morpheme realizes more than one regular case other than instrumental case
(i.e. nominative, accusative, ergative, dative and genitive).
The rest of this section is organized as follows. First, I will introduce the
case hierarchy (CH) (Silverstein 1980/1993; cf. Blake 2001; Caha 2013) and
will summarize what the CH has to say about the typological variation of
case syncretism among the regular cases. Second, I will illustrate the repre-
sentative types of syncretism predicted by the CH. Finally, I will show how
and to what extent these instances of case syncretism are derivable in terms
of OT.
(41a) states that if a language has a distinct morpheme for genitive, it has a
distinct morpheme for accusative and/or ergative, that if a language has a
distinct morpheme for accusative and/or ergative, it has a distinct mor-
pheme for dative and nominative/absolutive, while (41b) states that if a
language has a distinct morpheme for representing instrumental and/or
locative, it has a distinct morpheme for dative.
Five remarks are in order about (41). First, (41a) states that nominative/
absolutive and dative constitute the minimal case-marking system and that
(43) Kabardian (Colarusso 1992: 167; Smith 1996: 108, 111, 113)
a. ɬ’ -m
e š -r e f ́ z -m
e e j r jtáhs.
e e
˙
man-obl horse-nom woman-obl (nom.3).io.act.gave
‘The man gave the horse to the woman.’
b. ɬ’ -m
e š -r e j -w h’áhs.
e e
˙
man-obl horse-nom (nom.3)-act-killed
‘The man killed the horse.’
c. ha-r žás -m e mabáhna.
˘ ˙
dog-nom night-obl (nom.3)-bark
‘The dog barks at night.’
d. ɬ’ -r
e f ́ z -m
e e náxra nax’’ ́ zs. e
˙˙
man-nom woman-obl older (nom.3)-is
‘The man is older than the woman.’
e. máz -m e jahh.
˘
forest-obl act.(nom.3).carry
‘They carry it to the forest.’
f. ha-m Ø-y -pa-re
˘
dog-obl 3-poss-nose-nom
‘the dog’s nose’
The nominative case suffix -r marks S and O arguments, while the oblique
case suffix -m marks A arguments, a wide range of oblique core arguments
and adjuncts, and adnominal possessors. In contrast to the nominative case
suffix, which may appear only once per clause, the oblique case suffix may
appear multiple times within a clause. The fact that Kabardian distinguishes
A arguments from S and O arguments in terms of case marking suggests
that its case system (as well as its cross-referencing system) involves an
ergative alignment.
In contrast to Kabardian, Palauan involves an accusative case alignment.
(44a)–(44f) show that it uses the oblique preposition er to mark non-
macrorole core arguments, human and/or specific-and-singular
O arguments in imperfective clauses, adjuncts and adnominal possessors,
while leaving S and A arguments unmarked.27
(44) Palauan (Josephs 1975: 235; Georgopoulos 1991: 26, 27, 29)
a. ak-mo er a katsudo.
r.1sg-go obl movies
‘I am going to the movies.’
b. ng-kiltmekl-ii a ulaol a Peter.
r.3sg-clean-3sg floor Peter
‘Peter cleaned the floor.’
c. ng-diak ku-nguiu er a hong.
neg irr.1sg-read obl book
‘I am not reading the book.’
d. ng-mo er a ngebard er a klukuk.
r.3sg-go obl west obl tomorrow
‘She is going to America tomorrow.’
e. A Romana a omeka er a rengalek
Romana feed obl children
er a kukau.
obl taro
‘Romana is feeding the children the taro.’
f. ak-uleldanges er a resensei er ngak
r.1sg-honor.ipfv obl teachers obl me
‘I respected my teachers.’
NOM -r
ERG -m (oblique)
GEN
DAT
INSTR, LOC . . .
NOM ø
ACC er (oblique)
GEN
DAT
INSTR, LOC . . .
(45a,b) indicate that the same case morpheme is used to mark recipient
arguments of ditransitive verbs and adnominal possessors. This genitive-
dative syncretism renders (45c) ambiguous and allows the second-person
dative pronoun in (45c) to be construed as a benefactive or possessor (Tsu-
noda 1981: 110).
Likewise, Old Persian (Indo-Iranian) exhibits the same syncretism in its
accusative case system (Benvenuto and Pompeo 2015). Romanian (Romance)
provides another illustration of genitive-dative syncretism.
Case marking
NOM Nominative
ACC pe
GEN Oblique
DAT
Table 7.4 shows how undergoer arguments of transitive verbs are case-
marked in Estonian.
Furthermore, Finnish syncretizes genitive and accusative in the singular
declension, exhibits an aspectually conditioned case alternation on under-
goer arguments of transitive verbs (which is analogous to the one shown in
Table 7.4), and allows the genitive/accusative case to mark a wide variety of
nouns, including complement nouns of many postpositions and subject
The case morpheme that marks adnominal possessors and actor arguments
of transitive verbs in the Eskimo languages has been termed ‘relative case’ in
Eskimo linguistics. This and similar instances of the genitive-ergative syn-
cretism have repeatedly been argued to have arisen from (a reanalysis of )
nominalized constructions (Trask 1979; Bricker 1981; Starosta et al. 1982;
Johns 1992; Kaufman 2009; Coon 2009, 2013). The consensus in the literature
cited here is that the syntactic parallelism between the possessive and
transitive constructions goes a long way toward accounting for the
genitive-ergative syncretism.
Another example of the genitive-ergative syncretism comes from Tagalog,
which displays a symmetrical voice system with verbs bearing voice morph-
ology to indicate the semantic role of a nominative-marked argument, as
illustrated by (50a)–(50h).
(50) Tagalog (Kroeger 1993: 13, 14, 23, 32, 50; Sabbagh 2016: 658)
a. Pinutol ng¼magsasaka ang¼sungay ng¼kalabaw.
pfv.pv.cut gen¼farmer nom¼horn gen¼buffalo
‘The farmer cut off the buffalo’s horn.’ (Patient voice)
b. Ibinigay lahat ng¼mga¼guro sa¼mga¼bata
pfv.pv.give all gen¼pl¼teacher dat¼pl¼child
ang¼pera.
nom¼money
‘The teachers gave all the money to the children.’ (Patient voice)
c. Bumili ang¼lalake ng¼isda sa¼tindahan.
pfv.av.buy nom¼man gen¼fish dat¼store
‘The man bought fish at the store.’ (Actor voice)
d. Binili ng¼lalake ang¼isda sa¼tindahan.
pfv.pv.buy gen¼man nom¼fish dat¼store
‘The man bought the fish at the store.’ (Patient voice)
Four comments are in order about (50a)–(50h). First, (50a,c,d,e) show that
the genitive case marker ng/ni encodes not only adnominal possessors,
but also transitive actor arguments in non-actor (i.e. patient, benefactive,
locative or instrumental) voice constructions and transitive undergoer
arguments in non-patient voice constructions. Second, (50c) and (50d)
show that nominative-marked undergoer arguments of transitive verbs
in non-actor voice constructions are construed as specific (or generic),
while genitive-marked undergoer arguments of transitive verbs in actor
voice constructions are typically construed as non-specific.32 Third, Taga-
log requires undergoer arguments of transitive verbs to receive dative
case when they are pronouns or proper names that refer to animate
entities (as illustrated in (50f)).33 We may interpret the oblique marking
of Elias in (50f) as an instance of differential object marking, under the
assumption that sa syncretizes dative and accusative. Finally, Tagalog has
a voiceless construction with no overt dependency between the verb
inflected for recent-perfective aspect and one of its arguments: the verb
bears no voice morphology and neither of its arguments receives nom-
inative marking (Kroeger 1993: 50).34 (50g,h) show that single arguments
of intransitive verbs and two major arguments of transitive verbs receive
genitive case in recent-perfective constructions.35 This suggests that ng/ni
serves as the default case marker of core arguments (cf. Foley and Van
Valin 1984: 389).
Finally, the accusative-dative syncretism is found in a number of Indo-
Aryan languages (e.g. Hindi, Punjabi, Kashmiri) and Romance languages (e.g.
Spanish, Catalan, Southern dialects of Italian). For example, Hindi uses the
case clitic -ko to encode animate and/or specific undergoer arguments of
transitive verbs and recipient arguments of ditransitive verbs.
Likewise, Spanish uses the preposition a to mark not only recipient argu-
ments of ditransitive verbs and goal arguments of intransitive motion verbs,
but also undergoer arguments of transitive verbs that are typically both
human and specific.
Since Spanish has no genitive case morpheme in both nouns and pronouns,
it exhibits the two-way case-marking system (consisting of nominative and
dative) for nouns and local (first and second) person pronouns and the three-
way case-marking system (consisting of nominative, dative and accusative)
for third-person pronouns.36
(53) Northern Sámi (Aikio 2009; Valijärvi and Kahn 2017: 54–56)
a. Bussá čohkká stuolus.
cat.nom sit.prs.3sg chair.loc
‘The cat is sitting on a chair’ (Static location)
b. Bussá njuike stuolus láhttái.
cat.nom jump.prs.3sg chair.loc floor.ill
‘The cat jumps from the chair onto the floor.’ (Source)
c. Nieiddas lea ođđa irgi.
girl.loc be.prs.3sg new boyfriend.nom
‘The girl has a new boyfriend.’ (Possessor)
(53a)–(53e) show that the locative case conflates the distinction between
location and source in both the concrete local and possessive domains,
while the illative case marks goal arguments of motion verbs and recipient
arguments of ditransitive verbs. (53f,g) show that the locative case may
mark animate experiencer arguments and demonstrate that the locative
case has detached itself from the concrete locative system and has become
able to mark a non-spatial (as well as spatial) participant of an event.37
(53h) illustrates a further step taken by the locative case to move out of the
physical local domain, since it construes the action taken by Matthew as
involuntary or beyond his control (Aikio 2009).38 The fact that the locative
case conflates the location/source distinction and covers the dative domain
(Seržant 2015) except for recipient arguments and has been grammatica-
lized to be able to form a paradigmatic contrast with the nominative case
and to denote the low degree of agentivity suggests that the locative case
in Northern Sámi is abstracted from concrete locative domains to repre-
sent pure obliqueness, which the dative case in other languages is sup-
posed to represent.39
Second, the adessive case (one of the external local cases) in Estonian
encodes possessive and other non-spatial relations much more frequently
than spatial ones (Matsumura 1994); it marks possessor arguments, experi-
encer arguments, and causee arguments of causative verbs, and ‘logical
subjects’ of impersonal verbs, as illustrated by (54a)–(54i).40
(54) Estonian (Matsumura 1994: 225, 230, 231; Lindström and Vihman 2017: 2, 10)
a. Raamat on laua-l.
book.nom be.prs.3sg desk-ade
‘The book is on the desk.’ (Location)
b. Mu-l on uus auto.
1sg-ade be.prs.3sg new.nom car.nom
‘I have a new car.’ (Possessor)
The fact that the adessive case not only covers part of the dative domain (i.e.
possessor and experiencer arguments) but also causee arguments of transi-
tive verbs and ‘logical subjects’ of impersonal verbs (e.g. (54i)) leads Matsu-
mura (1994) to define the adessive as a functional equivalent to dative
(termed ‘adessive-dative’). The allative case also marks some experiencer
arguments as well as recipient arguments of ditransitive verbs, as illustrated
in (55a,b).
However, the crucial difference between the adessive and allative is that
experiencer arguments require or prefer the adessive marking when
accompanied by an infinitival complement (as illustrated by (54g)) (Lind-
ström and Vihman 2017).41 (54h,i) represents a further departure from
the locative use in (54a), since the adessive case in (54h,i) only marks the
highest-ranking direct core argument (‘logical subject’) of the embedded
infinitival verb. We may take these facts as an indication that the ades-
sive case in Estonian has been grammaticalized to be able to behave as
(56) Finnish (Sulkala and Karalainen 1992: 295; Huumo 1996: 81, 82; Karlsson 1999:
96; Kiparsky 2001: 356; Metslang and Erelt 2006: 263; Mazzitelli 2017: 28)
a. Poja-lla/*-n ol-i kirje.
boy-ade/gen be-pst.3sg letter.nom
‘The boy had a letter.’ (Possessor)
b. Poja-lta/*-n katos-i kolikko.
boy-abl/gen disappear-pst.3sg coin.nom
‘The boy lost a coin.’ (Deprived possessor)
c. Minu-n/lla ol-i nälkä.
1sg-gen/ade be-pst.3sg hunger.nom
‘I was hungry’ (Experiencer)
d. Minu-n/lla on ikävä hän-tä.
1sg-gen/ade be.prs.3sg sorry.nom he-prt
‘I feel sorry for him.’ (Experiencer)
e. Mies-ten on pakko poistua.
man-gen.pl be.prs.3sg obligation.nom leave.inf
‘The men have to leave.’ (Modal experiencer)
f. Mauno-n ol-i hauska pääs-tä kotiin.
Mauno-gen be-pst.3sg pleasant get-inf home.ill
‘It was nice for Mauno to get home.’
g. Vireni-n onnistui voittaa.
Viren-gen succeed.pst.3sg win.inf
‘Viren succeeded in winning.’
h. Anno-i-n sinu-n/Mati-n näh-dä karhu-n.
let-pst-1sg 2sg-gen/Matti-gen see-inf bear-gen
‘I let you/Matti see a/the bear.’
i. Pauli kirjoituttaa Harri-lla kirje-en.
Pauli.nom write.caus.prs.3sg Harri-ade letter-gen
‘Pauli makes Harri write a letter.’ (Causee)
j. Anno-i-n hei-lle karhu-n.
give-pst-1sg 3pl-all bear-gen
‘I gave them a bear.’ (Recipient)
Table 7.5 Dative domain in Northern Sámi, Estonian, Finnish and Tolmači Karelian
constructions has been replaced by the external local cases (including the
adessive case) and those uses of the genitive case that mark experiencer
arguments are competing with the adessive case (as in (56c,d)). (56h,i) show
that causees of transitive verbs receive genitive or adessive case, depending
on whether they occur in periphrastic or morphological causative construc-
tions.43 What is peculiar about Finnish is that the adessive case has been
replacing the ‘dative-genitive’ case, in contrast to Northern Sámi and Esto-
nian, which, respectively, use the locative and adessive case as some kind of
‘pure’ oblique marker.44
Finally, Tolmači Karelian, a dialect of Karelian (Finnic), uses the adessive
case to mark not only possessor and experiencer arguments but also recipi-
ent arguments of ditransitive verbs, due to the loss of the allative case
morpheme (which formerly marked goal and recipient arguments) (Oranen
2019). Oranen chooses to retain the traditional term ‘adessive’ (instead of
‘dative’) for etymological and historical reasons, but notes that the adessive
case in Tolmači Karelian is comparable to the dative case in Russian.45
To summarize, Northern Sámi, Estonian and Finnish use a different
case morpheme for recipient arguments of ditransitive verbs than the
one used for the rest of the dative domain (in contrast to Tolmači Kare-
lian, which uses the adessive case for the entire dative domain), but the
fact that the locative case (Northern Sámi), the adessive case (Estonian)
and the genitive or adessive case (Finnish) behave, to varying degrees, as a
‘pure’ oblique marker abstracted from concrete locative domains in a
paradigmatic contrast with the core cases suggests that the CH holds true
even in languages with no dedicated ‘dative’ case morpheme. Table 7.5
summarizes the data from the four languages that are discussed in
this subsection.
‘MAX [Case]’ requires each case feature value in the input (syntactic case) to
be realized by some case morpheme (morphological case) in the output,
while ‘IDENT [Case]’ requires the case feature values in the input not to be
different from the case value in the output. It is important to keep in mind
that ‘IDENT [Case]’ amounts to a local conjunction of ‘IDENT [Gen, Acc, Dat]’
and ‘IDENT [Gen, Acc, Dat],’ which means that ‘IDENT [Case]’ is satisfied
when a distinct accusative or ergative case morpheme is available in the
presence of a distinct genitive case morpheme.47 This is derived from (41a),
which requires a distinct accusative and/or ergative case morpheme in the
presence of a distinct genitive case morpheme.48
The third step is to derive the patterns of case syncretism in (42) from the
interaction of the markedness and faithfulness constraints in (57). Let us
begin with the massive syncretism in Kabardian, a Northwest Caucasian
language that involves an ergative case alignment. (58) is the constraint
ranking for Kabardian.
(58) *Gen >> {*Acc, *Erg} >> MAX [Case] >> IDENT [Case] >> *Dat
(58) ensures that the genitive and ergative case in the input are morpho-
logically realized by the dative case. This explains why actor arguments of
transitive verbs, adnominal possessors, and a wide range of oblique argu-
ments and adjuncts bear the same morphological case marking. Tableau
7.4 shows that the constraint ranking in (58) outputs the dative case
morpheme in Kabardian when it receives the syntactic ergative or genitive
case as input.
ˈˈ MAX IDENT
ˈˈ
*Gen *Acc
ˈˈ *Erg [Case] [Case] *Dat
ˈˈ
ˈˈ
GEN *!
ERG ˈˈ
ˈˈ
*!
☞DAT ˈˈ
ˈˈ
* *
NOM
ˈˈ *!
ˈ
Tableau 7.5 Case syncretism in Old Persian
MAX IDENT ˈˈ
ˈˈ
*Gen [Case] [Case] *Acc
ˈˈ *Erg *Dat
ˈˈ
ˈˈ
GEN *!
ACC ˈˈ
ˈˈ
* *!
☞DAT ˈˈ
ˈˈ
* *
NOM *! ˈˈ
ˈ
Re-ranking of ‘MAX [Case]’ and ‘IDENT [Case]’ as shown in (59) yields the
genitive-dative syncretism, illustrated by Old Persian.
(59) *Gen >> MAX [Case] >> IDENT [Case] >> {*Acc, *Erg} >> *Dat
(60) MAX [Case] >> IDENT [Case] >> *Gen >> *Acc >> *Erg >> *Dat
Given that ‘IDENT [Case]’ requires a distinct accusative and/or ergative case
morpheme in the presence of a distinct genitive case morpheme and that
faithful realization of accusative incurs a more serious violation than that of
ergative in (60), (60) leads accusative to be syncretized with dative, while
requiring ergative to be realized faithfully, as shown in Table 7.6.49
The next question to ask is how to derive the genitive-ergative syncretism.
It is important to note that no matter how the constraints in (57) may be
ranked, it is impossible to turn the genitive into the ergative. The reason is,
simply, that the dative case always beats the ergative case. This leaves us
with no choice but to postulate a case morpheme that underspecifies its case
value, so that it may be compatible with the genitive or ergative.50 The above
discussion suggests that ng/ni, sa/kay and ang/si constitute a case morpheme
with no inherent case value, a dative case morpheme and a nominative case
morpheme, respectively.
The major piece of evidence for analysing ng/ni as having no inherent case
value comes from the fact that they mark not only A, O and S arguments in
voiceless constructions but also some instrumental nouns, as illustrated by
(61a,b) (Kroeger 1993: 45).
The fact that ng/ni-marked nouns include, but are not restricted to, actor
arguments of transitive verbs and adnominal possessors demonstrates that
ng/ni lack any inherent case value and that ng/ni functions as the default
(elsewhere) case morpheme.51
An analogous account can be given of the genitive-accusative syncre-
tism in Northern Sámi, which allows its genitive/accusative case mor-
pheme to mark complement nouns of both postpositions and
prepositions, subject arguments of non-finite clauses, and nouns that
represent a standard of comparison in comparative constructions (see
Section 7.5.3) in addition to undergoer arguments of transitive verbs
and adnominal possessors.
(62a,b) and (47a)–(47c) (repeated above as (62c)–(62e)) show that the genitive/
accusative case morpheme in Northern Sámi behaves rather as the default
case morpheme and hence has no inherent case value of its own.52
The genitive-accusative syncretism also obtains in Estonian. The genitive/
accusative case morpheme in Estonian not only marks adnominal posses-
sors and some undergoer arguments of transitive verbs (see Section 7.5.2),
but also modifiers of adjectives, complement nouns of most postpositions
and some prepositions, and measure adverbials that are singular (Ehala
1994: 180–181; Miljan 2008: 176). The latter two uses are illustrated in
(63a)–(63d).
The case alternation in (63b) shows that measure adverbials that are plural
receive partitive case, while those that are singular receive genitive case.53
I follow Miljan and Cann (2013) in defining the genitive/accusative case in
Estonian as involving ‘non-subject’ dependency on some head (i.e. a verb,
noun or adposition) and deriving its contextual interpretation as found in
(63b) and (48b) (repeated as (63c), which is in contrast with (63d)) (quantitative
specification and aspectual boundedness) inferentially from its paradigmatic
contrast with the partitive case (which Miljan and Cann (2013) analyse as
carrying the meaning of ‘part of’; cf. Brattico 2011; Metslang 2014; Norris
2018). Their analysis of the genitive/accusative case morpheme in Estonian as
being ‘structural’ amounts to defining it as having no case value.
To summarize, this subsection has shown that there are two types of case
syncretism, one that involves markedness reduction and is derived from
7.6 Conclusion
This chapter has outlined the RRG theory of case assignment in its original
formulation (Van Valin 1991, 1993, 2005, 2009; Van Valin and LaPolla 1997),
its OT implementation (Nakamura 1999a, 1999b) and its extension to instru-
mental case assignment and case syncretism (Nakamura 2011, 2015, 2021).
The original RRG case theory is a version of dependent case theory that is
dependent not on phrase-structural asymmetry but on the ranking of actor
and undergoer, while its OT implementation not only defines nominative
and absolutive (conflated as the any-argument case despite their distribu-
tional difference), accusative and ergative case with no appeal to the
ranking of macroroles, but also derives the typological variation of case
systems from re-ranking of the four simple constraints.
Assuming either version of the RRG case theory, the chapter has also
outlined two attempts to further extend its scope: (i) providing the new
macrorole-based definition of instrumental case and deriving the diverse
interpretations of instrumental-marked nouns from the interaction of its
monosemous meaning with various contextual information rather than
from the alleged polysemy of instrumental case (Wierzbicka 1980; Janda
1993; Narrog and Ito 2007) and (ii) deriving the typological variation of case
syncretism among dative, accusative, ergative and genitive from the compe-
tition of the markedness constraints derived from the Case Hierarchy (Sil-
verstein 1976, 1980/1993) and the faithfulness constraints and the default
case morphemes with no case value (Nakamura 2002, 2011).
References
Durie, Mark. 1985. A Grammar of Acehnese on the Basis of a Dialect of North Aceh.
Dordrecht: Foris
Durie, Mark. 1987. Grammatical relations in Acehnese. Studies in Language 11:
365–399.
Ehala, Martin. 1994. Russian influence and the change in progress in the
Estonian adpositional system. Linguistica Uralica 3: 177–193.
Farrell, Patrick. 2009. The preposition with in Role and Reference Grammar.
In Lilián Guerrero, Sergio Ibáñez and Valeria A. Belloro( eds.), Studies in Role
and Reference Grammar, 179–202. Mexico City: UNAM Press.
Foley, William A. and Robert D. Van Valin, Jr. 1984. Functional Syntax and
Universal Grammar. Cambridge: Cambridge University Press.
Fowler, George. 1996. Oblique passivization in Russian. The Slavic and East
European Journal 40: 519–545.
Georgopoulos, Carol. 1991. Syntactic Variables: Resumptive Pronouns and A’
Binding in Palauan. Dordrecht: Kluwer Academic Publishers.
Gerdts, Donna B. 1988. Object and Absolutive in Halkomelem Salish. New York:
Garland.
Gönczöl-Davies, Ramona. 2008. Romanian: An Essential Grammar. London:
Routledge.
Hale, Ken. 1983. Warlpiri and the grammar of non-configurational lan-
guages. Natural Language and Linguistic Theory 1: 5–47.
Hill, Virginia. 2013. The direct object marker in Romanian: A historical
perspective. Australian Journal of Linguistics 33: 140–151.
de Hoop, Helen and Andrej L. Malchukov. 2008. Case-marking strategies.
Linguistic Inquiry 39: 565–587.
Huumo, Tuomas. 1996. Domain shifts and the grammaticalization of case:
A case study of the Finnish adessive. Folia Linguistica Historica 30: 73–96.
Inaba, Nobufumi and Rogier Blokland. 2001. Allative, genitive, and partitive:
On the dative in Old Finnish. Congressus Nonus Internationalis Fenno-Ugris-
tarum 7, Pars IV: 421–431.
Jakobson, Roman. 1936/1984. Beitrage zur allgemeinen Kasuslehre:
Gesamtbedeutungen der russischen Kasus. Travaux du Cercle Linguistique
de Prague 6: 240–288. Reprinted in Linda R. Waugh and Morris Halle
(eds.), 1984. Russian and Slavic Grammar: Studies 1931–1981, 59–103. Berlin:
Mouton.
Janda, Laura A. 1993. The Geography of Case Semantics: The Czech Dative and the
Russian Instrumental. Berlin: Mouton de Gruyter.
Johns, Alana. 1992. Deriving ergativity. Linguistic Inquiry 23: 57–87.
Jolly, Julia A. 1991. Prepositional Analysis within the Framework of Role and
Reference Grammar. New York: Peter Lang.
Jolly, Julia A. 1993. Preposition assignment in English. In Robert D. Van
Valin, Jr. (ed.), Advances in Role and Reference Grammar, 275–310. Amsterdam:
John Benjamins.
Josephs, Lewis S. 1975. Palauan Reference Grammar. Honolulu: University Press
of Hawaii.
Metslang, Helena. 2014. Partitive noun phrases in the Estonian core argu-
ment system. In Silvia Luraghi and Tuomas Huumo (eds.), Partitive Cases
and Related Categories, 177–256. Berlin: Mouton de Gruyter.
Metslang, Helle and Matti Erelt. 2006. Estonian clause patterns: From Finno-
Ugric to Standard Average European. Linguistica Uralica 42: 254–266.
Miljan, Merilin. 2008. Grammatical Case in Estonian. Unpublished PhD disser-
tation, University of Edinburgh.
Miljan, Merilin and Ronnie Cann. 2013. Rethinking case marking and case
alternation in Estonian. Nordic Journal of Linguistics 36: 333–379.
Mohanan, Tara. 1994. Argument Structure in Hindi. Stanford, CA: CSLI
Publications.
Nakamura, Wataru. 1999a. Functional Optimality Theory: Evidence from
split case systems. In Michael Darnell et al. (eds.), Functionalism and Formal-
ism in Linguistics, Vol. 2: Case Studies, 253–276. Amsterdam: John
Benjamins.
Nakamura, Wataru. 1999b. An Optimality-Theoretic account of the Japanese
case system. Studies in Language 23: 607–660.
Nakamura, Wataru. 2002. Deriving morphological cases: Markedness consid-
erations. In Reinhard Rapp (ed.), Sprachwissenschaft auf dem Weg in das dritte
Jahrtausend. Akten des 34. Linguistischen Kolloquiums in Germersheim 1999, Teil
1, 151–158. Frankfurt am Main: Peter Lang.
Nakamura, Wataru. 2008. Fluid transitivity and generalized semantic roles.
In Robert D. Van Valin, Jr. (ed.), Investigations of the Syntax–Semantics–
Pragmatics Interface, 101–116. Amsterdam: John Benjamins.
Nakamura, Wataru. 2011. Case syncretism in typological perspective: An
RRG-OT account. In Wataru Nakamura (ed.), New Perspectives in Role and
Reference Grammar, 35–63. Newcastle upon Tyne: Cambridge Scholars
Publishing.
Nakamura, Wataru. 2015. A neo-Jakobsonian account of default oblique
cases: Instrumental vs. dative. Paper read at the 2015 International Con-
ference on Role and Reference Grammar, Heinrich Heine University
Düsseldorf.
Nakamura, Wataru. 2016. A two-tiered theory of case features: The case of
the Hindi case(-marking) system. Paper presented in the 49th Annual Meet-
ing of the Societas Linguistica Europaea, University of Naples Federico II.
Nakamura, Wataru. 2021. A neo-Jakobsonian account of default oblique
cases: Instrumental vs. dative. In Robert D. Van Valin, Jr. (ed.), Challenges
in the Analysis of the Syntax–Semantics–Pragmatics Interface. Newcastle upon
Tyne: Cambridge Scholars Publishing.
Narasimhan, Bhuvana. 1998. A lexical-semantic explanation for ‘quirky’ case
marking in Hindi. Studia Linguistica 52: 48–76.
Narrog, Heiko and Shinya Ito. 2007. Re-constructing semantic maps: The
comitative-instrumental area. Sprachtypologie and Universalienforschung 60:
273–292.
Starosta, Stanley, Andrew K. Pawley and Lawrence A. Reid. 1982. The evolu-
tion of focus in Austronesian. In Amram Halim, Lois Carrington and S. A.
Wurm (eds.), Papers from the Third International Conference on Austronesian
Linguistics, Vol. 2: Tracking the Travelers, 145–170.
Stolz, Thomas, Cornelia Stroh and Aina Urdze. 2006. On Comitatives and
Related Categories: A Typological Study with Special Focus on Languages of Europe.
Berlin: Mouton de Gruyter.
Sulkala, Helena and Merja Karalainen. 1992. Finnish. London: Routledge.
Tamm, Anne. 2007. Perfectivity, telicity, and Estonian verbs. Nordic Journal of
Linguistics 30: 229–255.
Tomić, Olga Mišeska. 2004. The Balkan Sprachbund properties: An introduc-
tion. In Olga Mišeska Tomić (ed.), Balkan Syntax and Semantics, 1–55. Amster-
dam: John Benjamins.
Trask, Robert L. 1979. On the origins of ergativity. In Frans Plank (ed.),
Ergativity: Toward a Theory of Grammatical Relations, 385–404. London: Aca-
demic Press.
Tsunoda, Tasaku. 1981. The Djaru Language of Kimberley, Western Australia.
Canberra: Pacific Linguistics.
Valijärvi, Riita-Liisa and Lily Kahn. 2017. North Sámi: An Essential Grammar.
London: Routledge.
Van Valin, Robert D., Jr. 1991. Another look at Icelandic case marking and
grammatical relations. Natural Language and Linguistic Theory 9: 145–194.
Van Valin, Robert D., Jr. 1993. A synopsis of Role and Reference Grammar. In
Robert D. Van Valin, Jr. (ed.), Advances in Role and Reference Grammar, 1–164.
Amsterdam: John Benjamins.
Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cam-
bridge: Cambridge University Press.
Van Valin, Robert D., Jr. 2009. Case in role and Reference Grammar. In
Andrej Malchukov and Andrew Spencer (eds.), The Oxford Handbook of Case,
102–120. Oxford: Oxford University Press.
Van Valin, Robert D., Jr. 2012. Lexical representation, co-composition, and
linking syntax and semantics. In James Pustejovsky, Pierrette Bouillon,
Hitoshi Isahara, Kyoko Kanzaki and Chungmin Lee (eds.), Advances in
Generative Lexicon Theory, 67–107. Dordrecht: Springer.
Van Valin, Robert D., Jr. 2013. Head-marking languages and linguistic theory.
In Balthasar Bickel, Lenore A. Grenoble, David A. Peterson and Alan
Timberlake (eds.), Language Typology and Historical Contingency, 91–123.
Amsterdam: John Benjamins.
Van Valin, Robert D., Jr. 2018. Some issues regarding (active) accomplish-
ments. In Rolf Kailuweit, Lisann Künkel and Eva Staudinger (eds.), Applying
and Expanding Role and Reference Grammar, 71–93. Freiburg: Albert-Ludwigs-
Universität, Universitätsbibliothek.
Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning,
and Function. Cambridge: Cambridge University Press.
Notes
1 See Levin and Rappaport Hovav (2005: 51–68) for a critical comparison of
macroroles with proto-roles (Dowty 1991).
2 The underlying assumption is that the relative degree of affectedness
determines which non-actor argument serves as undergoer. See Van Valin
(2005: 109–115) for further discussion.
3 Van Valin (1991, 2005) attributes the idea of dative being the default case
to Silverstein (1980/1993).
4 For example, Japanese allows source arguments of ditransitive verbs such
as morau ‘receive’ and kariru ‘borrow’ to be marked by the dative (as well as
the ablative) case particle.
5 See Jolly (1991: 115–119) for an LS-based account of some other non-
locative uses of from (e.g. John suffered from arthritis).
6 German behaves like Icelandic with respect to case preservation under
passivization, while Japanese allows no such case preservation. See
Section 7.4.1 for an OT reformulation of the case assignment rules in (9)
and how it accounts for the contrast between Icelandic/German and
Japanese with respect to case preservation.
7 Predicate adjectives that modify covert controllee arguments of depend-
ent cores in Icelandic control constructions may receive nominative case
or the case of the controller argument in the matrix core (Andrews 1982:
450–456). This fact shows that the controllee argument may receive nom-
inative case and agree in case with predicate adjectives that modify it. We
may take this agreement fact as another piece of evidence that the DCA
for Icelandic is the core.
8 Matrix-coding constructions are also attested in ergative and active lan-
guages. For example, Basque requires actor arguments of transitive verbs
in non-finite gerundive complements of perception verbs to receive abso-
lutive case (Rezac et al. 2014).
9 The auxiliary verb in (24b) (d-u-ø) contains the absolutive affix d-, which
refers to no clausal argument. The verb dantzatu is a denominal verb that
incorporates dantza ‘dance’ and the absolutive affix refers to this incorp-
orated noun. See Preminger (2012) for related discussion.
10 Aldai’s argument is based on the case marking of S arguments of lexically
simple unergative verbs. See Aldai (2009: 797–800) for discussion of com-
pound unergative predicates formed with egin ‘do, make’ and a few other
light verbs. The fact that Basque has an ergative dialect in addition to an
active dialect justifies analysing the actor and undergoer markers in the
Western dialect as ergative and absolutive, respectively.
11 See Farrell (2009) for an exception to the ICR.
12 It is questionable whether (28) is in a subset relation with (9c). Rissman
(2010) compares the locatum with-phrases with the instrumental with-
phrases and shows that the former are syntactic arguments, while the
latter are adjuncts. See also Koenig et al. (2008) for related discussion of
the semantic and syntactic status of implement with-phrases.
13 Van Valin and LaPolla (1997: 381–382) attempt to extend the ICR to the
uses of with that mark manner adverbials in English (e.g. with enthusiasm,
with difficulty).
14 See de Hoop and Malchukov (2008) for an alternative OT-based case theory.
15 Russian allows some instrumental and genitive-marked arguments (but
not any dative-marked arguments) to receive nominative case under
passivization (Fowler 1996). One may be tempted to interpret this fact
as an indication that Russian also belongs to (33b) (under the assumption
that non-subject genitive/instrumental-marked arguments are not
undergoers), but this move would leave it unexplained why dative-
marked arguments may not undergo passivization in Russian.
16 It remains an open question whether or not there are any other non-
accusative case systems that do not allow case preservation under
passivization.
17 An asterisk mark (*) indicates a violation incurred by each candidate for a
given constraint. The asterisk followed by the exclamation mark indi-
cates a fatal violation and the candidate that incurs that violation is
eliminated from further consideration. The shaded cells indicate that
they exert no influence on the selection of the optimal output(s).
18 What is termed ‘instrumental’ in (35b) includes the ablative case in
Latin, which combines three formerly distinct cases, instrumental, loca-
tive and ablative proper (Oniga 2014: 254).
19 See Van Valin (2005: 111–112) for more relevant data from Croatian
and Dyirbal.
20 The instrumental adposition/case has no LS of its own and therefore has
no argument slot for a noun it marks (Van Valin and LaPolla 1997; Van
Valin 2005).
21 The notion of QS was incorporated into RRG by Van Valin and
LaPolla (1997).
22 The fact that the possessive relation between Petr and his body part is
permanent and holds outside the event of nodding indicates that the
main verb kivnutʹ ‘nod’ and covert possessive predicate do not form a
complex nucleus.
23 (34c) is comparable to secondary predicate constructions, which describe
the (resulting) state of a clausal participant and involve a non-
subordinate core juncture (Schultze-Berndt and Himmelmann 2004).
They share the same juncture–nexus type, but the difference lies in the
fact that the possessum noun in (34c) specifies which part of the posses-
sor is directly involved in the event instead of representing a temporary
state of the possessor.
24 We may compare the two types of dative in (41a,b) to the two types of the
dative case particle ni in Japanese (Sadakane and Koizumi 1995): ni as a
(non-predicative) case marker and ni as a predicative postposition. See also
Blansitt (1988), which may be taken as an attempt to elaborate (41b) by
establishing implicational relations among dative, allative and locative.
25 Silverstein (1980/1993) draws an explicit analogy between the develop-
ment of case-marking systems and that of the colour-term systems
(Berlin and Kay 1969; Kay and Maffi 1999).
26 The genitive case clitic in Hindi may case-mark subjects of a small
number of intransitive verbs (Mohanan 1994) despite its adjective-like
declension (it agrees in gender, number and case with a possessum noun
it occurs with) (Agnihotri 2007: 187–189). The chimeric status of this
genitive clitic is indicated in (42i) by putting ‘Gen’ in parentheses.
27 The morpheme a in (44) is an NP marker and is not glossed (Georgopou-
los 1991: 237).
28 Personal pronouns preserve a four-way case distinction (i.e. nominative,
accusative, dative and genitive) (Gönczöl-Davies 2008). We can derive the
lack of accusative marking on some transitive undergoer arguments
from the definiteness/animacy-based split accusativity.
29 Third-person agreement markers in matrix indicative clauses alone oper-
ate on an ergative-absolutive basis (Wiltschko 2006).
30 For simplicity, in this paper, I gloss the genitive/accusative case mor-
phemes in the examples taken from Northern Sámi, Estonian and Finn-
ish as ‘genitive.’
31 See Tamm (2007) for a detailed semantic account of the alternation
between the partitive and accusative/nominative case on undergoer
arguments of transitive verbs. See also Miljan and Cann (2013: 365–372)
for how the notion of aspectual boundedness arises.
32 The genitive case marker turns out to be neutral to the distinction
between specific and non-specific, since it may mark specific common
nouns and pronouns in voiceless constructions, as illustrated by (50g,h).
33 See Latrouite (2011: Ch. 5) and Sabbagh (2016: 656–660) for further
discussion of the clause-internal distribution of ng/ni and the case alter-
nation between ng/ni and sa/kay.
34 Kroeger (1993: 50) notes that gerundive constructions also bear no voice
marking on the verb stem and allow only what he terms ‘default
case assignment’.
35 Recent-perfective constructions are a type of impersonal (or ‘pivotless’)
construction. See Kroeger (1993: 48–51) for a list of impersonal construc-
tions in Tagalog.
36 An ergative-dative syncretism is hardly attested despite the fact that it is
predicted to be possible by the CH. The cross-linguistic paucity of the
ergative-dative syncretism requires a principled explanation. Nakamura
(2011) attributes its near unavailability to an independent semantic con-
straint to the effect that the farther one of the co-arguments of a verb is
away from the other one on the causal chain, the less likely they are to
receive the same morphological marking. The essential idea is that the CH
works in tandem with this semantic constraint (adapted from Croft (1991:
Ch. 5)) to constrain the possible patterns of case syncretism.
37 Aikio (2009) notes that Máhtes in (53g) does not serve as an
adnominal possessor.
38 Some verbs subcategorize for a locative-marked argument that resists
any immediate spatial interpretation (e.g. heaitit ‘stop’, dolkat ‘get sick of’,
beroštit ‘care about’). The same holds for verbs with an illative-marked
argument (Valijärvi and Kahn 2017: 54, 56).
39 (53h) is a crucial example, since it shows that the locative case can be
treated on a par with the nominative case for the purpose of indicating
the relative degree of agentivity.
40 The external local cases (allative, adessive and ablative) have a much
stronger tendency to mark non-spatial relations than the corresponding
internal local cases (illative, inessive and elative) in Estonian. As a case in
point, Matsumura (1994) compares the adessive case with the inessive
case in terms of the frequencies of three semantic types, (i) place, (ii)
time, and (iii) possessor, in two text corpora and finds that the third type
accounts for more than 40% of all the adessive nouns, while the first type
that refers to a physical space accounts for less than 8%. In contrast, the
first type accounts for almost 80% of all the inessive nouns, while no
inessive noun of the third type (possessor) is found.
41 See Matsumura (1996: 40–51) and Lindström and Vihman (2017: 9–11) for
more relevant examples and further discussion of the relation between
the adessive and allative in Estonian.
42 The allative, genitive and partitive cases mark recipient arguments of
ditransitive verbs in Old Finnish (Inaba and Blockland 2001).
43 The fact that the genitive marking of sinu-n/Mati-n in (56h) is retained
under passivization and negation (Kiparsky 2001: 356–357) indicates that
they are non-macrorole core arguments.
44 Korhonen (1991/1996) notes that the adnominal uses of the genitive case
were historically derived from the Proto-Uralic -n lative case by way of its
dative (‘dative-genitive’) uses. See also Huumo (1996: 74–75, 83–86) for
the reason the functions borne by the -n lative case have been taken over
by the external local cases.
45 Oranen (2019: 211–219) lists the functions borne by the adessive case
in Tolmači Karelian and compares its adessive case with the Finnish
and Estonian counterparts. See also Koivisto (2017: 162–164) for other
syncretic patterns (the adessive-ablative syncretism and allative-adessive-
ablative syncretism) displayed by a few other dialects of Karelian.
46 (57a)–(57c) make no reference to the nominative case value, but it is not
representationally underspecified. I follow the spirit of Calabrese (1995)
in parameterizing rules/constraints with respect to whether they may
refer to (i) all, (ii) contrastive, or (iii) marked feature values and assume
that all constraints related to case assignment may refer to contrastive
case values. It follows, then, that the nominative case value remains
invisible to constraints when it is not contrastive in the case-marking
system of a particular language (cf. Nevins 2007: 285–287).
47 Unlike ‘IDENT [Case]’, ‘MAX [Case]’ involves no local conjunction (as
shown in (57c)), since it is not sensitive to the distinctness/identity of
case morphemes and only ensures that a syntactic non-nominative case
is realized by any non-nominative case morpheme.
48 See McCarthy (2008: 214–219) for a formulation and discussion of local
conjunction (due originally to Smolensky (1995)).
49 The distinction between ng and ni is that the former marks common
nouns, while the latter marks personal names. The same contrast obtains
in sa/kay and ang/si.
50 The case-marking system in Hindi is derived from the same constraint
ranking as (60) (Nakamura 2016). The difference between Tagalog and
Hindi comes down to the availability of a case morpheme that neutral-
izes the ergative-genitive distinction.
51 This formulation accords well with Foley and Van Valin’s (1984: 389) defin-
ition of ng/ni as the default case marker for non-pivot core arguments.
52 Like Northern Sámi, Estonian uses both postpositions and prepositions,
but the latter are a relatively new development and are limited in
number. Specifically, the vast majority of postpositions in Estonian
assign genitive case to their complement nouns. In contrast, the case
assignment of prepositions is not dominated by any single case, but the
genitive case outnumbers the other cases (Ehala 1994: 180–181).
53 Measure adverbials as illustrated in (63b) exhibit the same case-marking
pattern as undergoer arguments of transitive verbs. See Maling (1993),
Nelson (1998, 2003) and Kiparsky (2001) for further illustration and
discussion of case-marked adverbials (like the one in (63b)) in the Finnic
and Sámi languages. See Van Valin (2018) for how to represent analogous
measure adverbials in terms of LS.
Francisco J. Cortés-Rodríguez
8.1 Introduction
The synthetic approach has to do with the mechanisms that are invoked
by morphological models to account for the production of morphologically
complex words. From a constructivist perspective, analytical morphology
takes the perspective of the decoder, whereas when a speaker produces a
new complex word they make use of the synthetic machinery.
From the point of view of theory building and description, the analytical
approach must come first. Only after a morphological theory has established
what the building blocks of words are will it be in a position to give an
account of the manner in which such pieces are put to work to create
complex lexical units. In line with this, our description of morphological
theory within RRG will also start with an overview of the proposals offered
within the model to provide a sound description of the internal constituents
of words. This is the main goal of Section 8.2 in this chapter. Based on the
programmatic contribution by Everett (2002) and the developments into a
fully articulated model in Martín Arista (2008, 2009, 2011), we will spell out
the details of the layered structure of the word (henceforth LSW). Some
alternative views on specific aspects of Martín Arista’s proposal (such as
those expressed in Cortés-Rodríguez and Sosa 2012; Nolan 2011; Diedrichsen
2011) will also be considered.
The synthetic aspects of morphological processes are considered in
Sections 8.3 and 8.4. From a synthetic viewpoint, attention is paid to issues
such as whether there are specific morphological rules to generate complex
words or, on the contrary, they are the result of other types of processes, as
are syntactic or phonological operations. Consequently, in synthetic descrip-
tions, one must take into account whether morphology is a separate com-
ponent of grammar or just a set of phenomena assimilated to other
modules. The functional orientation of RRG has led researchers interested
in this area to place the emphasis on the external motivation of morpho-
logical phenomena. In this regard, there is widespread consensus among
RRG morphologists that lexis, syntax, semantics and pragmatics motivate
morphology.1
It is at this point, however, that inflectional and derivational morphology
take different paths, inflectional morphology aligning itself with syntactic
processes while word-formation processes are treated as essentially lexical-
semantic phenomena. Following Stump’s (2001) four-way classification of
morphological theories, inflectional morphology in RRG follows an
inferential-realizational approach. In opposition to lexical theories of inflec-
tion, in which inflectional affixes have lexical entries specifying their mor-
phosyntactic properties, an inferential theory associates morphosyntactic
properties with morphological rules, which relate a given inflected word-
form to its root. Furthermore, instead of being incremental, the RRG
approach to inflection is realizational: the association of a given root or
lexeme (e.g. work) to a set of properties like ‘3rd person singular’, ‘present
tense’ and ‘indicative mood’ licenses the attachment of the affix -s (cf. Stump
2001: 1–3).2 Derivational morphology lies outside this typology, as it is, in
This section will establish the set of common assumptions that support the
different contributions for the development of morphology within RRG
from an analytical viewpoint. The description will proceed as follows.
Section 8.2.1 lists the basic underpinnings of the morphological theory
outlined in this chapter. Section 8.2.2 outlines the layered structure of the
word, the analytical tool responsible for grammatical analyses within the
word domain in RRG. Finally, Section 8.2.3 sketches the state of research on
operators in the LSW.
(a) There are two types of lexical morphemes (or lexemes): free and
bound lexemes. Free lexemes belong to the major lexical categories,
whereas bound lexemes are derivational affixes. Both free and bound
lexemes will be semantically represented by means of their corres-
ponding lexical entry. Both are also grouped within the lexicon into
lexical classes defined by their similarity of meaning. The difference
between these types of lexeme lies only in their distributional
behaviour.4
(b) It follows from what is outlined in (a) that affixal derivation and
compounding are essentially the same type of phenomenon. Despite
this, in languages like Spanish or English, there are some differences
between these two types of process. Whereas the grammatical
makeup of affixal formations is quite predictable, since derivative
affixes usually specify both the type of base they are attached to
and the type of lexical category that outputs from its combination,
‘compounding is only barely syntactic’ (Jackendoff 2009: 14), since
the elements of a compound are rather opaque, in that they do not
provide this kind of specification. This is especially clear in Spanish
compounds, which can have as their nucleus either the rightmost
(as in puti-club ‘whorehouse’, lit. ‘hooker-club’) or the leftmost
constituent (as in hombre-anuncio ‘sandwich-board man’, lit. ‘man-
advertisement’).
(c) Inflectional affixes are treated differently, as they are considered to be
morphemes with no lexical status (i.e. they are not lexemes) and
therefore they are not stored in the lexicon. Apart from Nolan
(2011: 74–75), for whom inflectional affixes are allocated to a ‘mor-
pheme store’, other researchers working in RRG morphology (Everett
2002; Martín Arista 2008; Boutin 2011; Van Valin 2013) argue for an
inferential realizational approach to inflection (cf. Spencer 1998,
2004; Stump 2001): phonological operations/rules will be responsible
for providing the morphosyntactic representation of a given inflected
form.5
(d) Whereas derivation and compounding are unified, inflection and word
formation are regarded as disparate phenomena from a functional point
of view, since they serve different purposes within the overall structure
of our grammatical model. Despite this, they must be treated together at
some point, as they contribute to the formal structure of the word,
which is the maximal unit in morphology. The main theoretical con-
struct that handles the overall internal makeup of the word, and which
consequently accounts for the structural distribution of inflectional and
derivational morphology, is the layered structure of the word (LSW).
Parallel to clauses and phrases, the internal structure of words is based
on layering. The first outline of the LSW can be found in Everett (2002),
and it becomes a fully articulated model in Martín Arista’s studies (2008,
2009, 2011).
WORD
CORE
NUCLEUS
PREDICATE
Figure 8.1 The layered structure of the word (Martín Arista 2009: 91)
WORDN
COREN
NUCN
friends'
NUC
CORE Number
WORD Case
COMPLEX WORDN
COREN
WORDN WORDN
COREN COREN
NUCN NUCN
NUC
CORE Number
COMPLEX WORDV
COREV
ARGN NUCLEUSV
WORDN PREDICATEV
COREN
NUCN
PREDN
COMPLEX WORDN
takal-takal mananyi
Figure 8.4 The LSW of a recursive complex word (adapted from Martín Arista 2009: 110)
the core and the word lexical operators, as we see above in Figure 8.2 for
friends’. From the analysis offered in Figure 8.3 it seems clear that among the
nominal inflectional features, number (and probably gender) are core oper-
ators, whereas case has scope over the word node.8
Nuclear operators, on the other hand, will encode those features that are
not subject to percolation into higher structures (i.e. syntax-blind features).
Thus, good candidates to be classified as nuclear operators are aspectualizers
in languages where lexical aspectual features are regularly marked by
morphological means, as is typical of Slavic languages.9 Other possible
nuclear operators are some types of lexical distributive processes in Amele
(Papua New Guinea) which involve a reduplication process, as in the
following clauses (data from Roberts 2015: 19–20):10
Roberts (2015: 19) explains that this type of reduplication takes place within
the verbal nucleus, in contrast with other types of reduplication processes
that take place outside the nucleus in higher layers, as shown by the
different distribution of grammatical markers.11
In (3) each verb has a different subject (DS) marker and 3SG subject
agreement. This is followed by verb inflection which includes 3PL agreement
with the matrix clause subject, mel ‘boys’, as well as present tense.
Table 8.1 shows a classification of the operators mentioned in the previous
description; it must be stressed that it is not an exhaustive classification and
that this is an issue that still needs further research.
(8) nus?ūlχ-Ø ‘
ti-gs-tx
thief-3sg.sbj prox-ill-prox
‘The one who is ill is a thief’
In Bella Coola, the attachment of person, number and case affixes, on the
one hand, and proximal morphemes, on the other, is flexible with regard to
class form, and is instead conditioned by the function of words in a given
sentence (Van Valin 2008). This type of phenomenon seems to run counter to
the establishment of a typologically valid arrangement of inflectional oper-
ators in the LSW (see Aronoff and Fudeman 2011: 201–202 for further
arguments on this); this is an issue which calls for further research.
(9) tsún-tsi-lhkan
tell-2sg.obj-1sg.sbj
‘I told you’
percolation across the LSW and the other grammatical domains is signifi-
cantly different if we compare head- vs. dependent-marking structures.
Whereas in the latter type of language, percolation is restricted to operators
across different structures (see Section 8.3.1), in head-marking structures,
morphological features percolate up to the constituent projection in the LSC
(cf. Section 8.3.2).
RP
CORE
LSP
(Constituent Projection)
NUC
WN = NOUN
COREW
NUCW
LSW
W INFL percolation
NUCW
COREW
WN = NOUN
NUC
RP Case
Figure 8.5 Feature percolation in RPs (adapted from Martín Arista 2008: 136)
SENTENCE
CLAUSE
CORE
RP NUC RP
WV = VERB
COREW
NUCW
NUCW
LSW
COREW imperf
WV =VERB past
WV =VERB person
WV =VERB number
NUC aspect
CORE
CLAUSE
tense
COREW
SENTENCE
CLAUSE
CORE
COREV
ACTOR UNDERGOER
representation of the core of the clause. In fact, the structure of the lexical
unit in the LSW will provide the structure of the core of the clause in the LSC;
Van Valin (2013: 115) describes that the two structures are coextensive, since
the structure of the COREw provides the structure of the core of the clause.
Note that the analysis of the COREw structure in Figures 8.7 and 8.8 is
almost identical to the structure of the core in the equivalent English clause
‘I gave it to them’. This proposal thus manages to provide comparable
SENTENCE
CLAUSE
CORE
NUC RP RP
WV =VERB
COREW
NUCW
NUCW
COREW imperf
WV =VERB past
WV =VERB person: 3
WV =VERB number: PL
NUC aspect
CORE
CLAUSE tense
Figure 8.9 The LSC of Spanish sentence cantaban villancicos
the LSW to both the operator and the constituent projections in the LSC of
the Spanish clause cant-aba-n villancicos (sing-ipfv.pst-3pl carols, ‘they were
singing carols’).
Even though Spanish is mainly a dependent-marking language, it shows
subject marking on the verb by means of person and number inflectional
affixes. This makes unnecessary the presence of an independent subject
constituent, as happens in this sentence, thus allowing head-marked struc-
tures. In the absence of an overt subject argument (e.g. los niños ‘the
children’ in Figure 8.6), the suffix -n is mapped from the LSW of the verb
into the constituent projection of the LSC; this is indicated by the dotted
line in the analysis. The other morphological features in the LSW of cant-
aba-n (sing-ipfv.pst-3pl), namely imperfective and past as encoded in the
suffix -aba, percolate into the operator projection of the LSC (cf. also
Figure 8.6).
This section deals with the other basic domain of morphology, namely the
creation of new lexical units. It is important to note that more often than
not grammatical models tend to focus their interest on the semantic repre-
sentation of clauses and phrases, but they usually leave aside the question of
how to account for the semantic representation of other grammatical struc-
tures such as compound and derived words. Contrary to this tendency, the
model outlined in this section offers a design for the semantic representa-
tion of such units which complies with the following issues: firstly, it
captures the semantic variability of word-formation patterns. The introduc-
tion of a two-tier enriched representation of lexical structures, which
includes not only the event features encoded in logical structures, but also
encyclopedic knowledge as captured by Qualia Theory, places RRG in an
outstanding position when compared with other proposals of derivational
morphology. Section 8.4.1 deals with these aspects and Section 8.4.2 offers
some illustrative analyses of the most relevant affixal units in Spanish and
English. Secondly, such a format of semantic representation will allow us to
account for the grammatical relations that hold between the components of
a derived lexeme within its corresponding LSW. This will be the aim of
Section 8.4.3.
These two aspects reveal a conception of derivational morphology as a
double-sided phenomenon: as a lexical process, word-formation involves
the creation of a new semantic structure in the wrapper of a lexical
unit; at the same time, the constituents of a derivationally complex
lexical item are interrelated by different types of grammatical relations,
that is, word formation is also a syntactic phenomenon within the
word domain.
lexicon). The lexical entries for both types of lexeme include a formal
semantic representation in the format of lexical templates. The latest
version of lexical templates for free lexemes (Cortés-Rodríguez 2009;
Mairal Usón, Periñan-Pascual and Pérez Cabello de Alba 2012; Ruiz de
Mendoza 2013) takes as a backbone for lexical representation the Aktion-
sart properties of predicates as formally encoded in the logical structures
of RRG and enriches them by integrating Pustejovsky’s (1995) Qualia
Theory together with the set of lexical functions from the explanatory
and combinatorial lexicology (Mel’cuk 1989; Mel’cuk and Wanner 1996;
Alonso and Tutin 1996). Despite their complexity, lexical templates are
fully fledged repositories of the semantic features associated with a
lexeme, either free or bound. By way of example, the lexical template
for the Spanish verb captar ‘fathom’ would have the following format
(Cortés-Rodríguez and Sosa 2012: 22):
(10) captar:
eventstr: know′ (x, y)
qualiastr: {Qf: manner : MagnObstr think′ (x, y)
Qt: Culm know′ (x, y <all>)}
The format of a lexical entry now consists of two basic components: the
event and the qualia structure. In this case, since captar belongs to the class
of Cognition verbs, the event structure is a state logical structure which
takes know′ as a primitive and has two arguments (x, y), following the
format of the logical structures of RRG. The second part of the lexical
template includes its qualia structure. Qualia Theory is borrowed from
Pustejovsky’s (1995, 2001) semantic model. Van Valin and LaPolla (1997:
184–186) already incorporated qualia as part of the semantic representation
of nouns. Our proposal takes it beyond this and aims at including it as a part
of any lexical semantic representation. What qualia tell us about a concept
is the set of semantic constraints by which we understand a word when
embedded in the language, which is also the purpose of our lexical tem-
plates. As Pustejovsky and Batiukova (2019: 162) state:
The core notion of qualia structure is that there are four aspects that
make up our knowledge of a word: the class of entities it denotes (the
formal role or quale), how the denoted entity was created (the agentive
role), the intended function of this entity (the telic role), and the
internal makeup of this entity (the constitutive role).
that exhibited by words in the primary lexicon. Some units are more
central or higher in the structure whereas other affixes are peripheral,
and some are even located in between two lexical classes. In this section a
brief description of some of the most relevant affixal classes in Spanish and
English will be offered (for a more detailed account, see Cortés-Rodríguez
2006a, 2006b, Cortés-Rodríguez and Sosa 2008, 2012). More specifically, an
account of the semantic aspects of one important group of nominaliza-
tions is given first, followed by an explanation of the intricate meaning
relationships between the components of derived causative
verbal formations.
This formula reads as follows: Any derived nominal lexeme φiN belonging to
this class denotes a participant entity xi of one of the events encoded in the
qualia structure (Qualia:…xi) that forms part of the lexical template of the
base lexeme LT…φBASE; hence the derived lexeme and that entity are co-
indexed by a superscript i. Another way to express this is to describe these
processes as nominalizations oriented towards an argument in the semantic
representation of the lexical unit that functions as the base of the deriv-
ation. There is an additional restriction on this entity: it must have macro-
role status. Therefore, the structure in (12) represents the ‘event structure’ of
the derivational class of concrete nominalizing affixes, or concrete proces-
sual substances/things/essences in Lieber’s (2004: 36) terminology. We propose
(13) φiN: [LT (xi. . ., [(e2: [LT. . .φBASE: (Qualia:. . .xi)])])], x ¼ actor
E.g. escritor ‘writer’, poseedor ‘owner’, panfletista ‘pamphleteer’, desodorante
‘deodorant’, lechero ‘milkman’.
(14) φiN: [LT (. . .xi, [(e2: [LT. . .φBASE: (Qualia:. . .xi)])])], x ¼ undergoer
E.g. arrendatario ‘tenant’, fideicomisario ‘trustee’, fallecido ‘deceased’
The labels ‘actor and undergoer nominalizations’ explain the wide scope
of these types of derivational processes. The semantics of the templates
cannot be associated with a specific semantic function such as ‘agent’.
Even though the most prototypical formations correspond to agent nom-
inals, like English writer, runner, violinist, etc., there are many other forma-
tions where the notion of ‘agenthood’ is absent (cf. formations like believer,
owner, lover, to mention just a few). The term ‘actor’ indicates that all the
formations are nominalizations of the macrorole actor, as defined within
RRG (see Chapter 4). This, in turn, justifies the superscript i which co-
indexes the lexical variable for the derived word (φN) with the participant
that would receive that macrorole function. That is, they mark the nom-
inals as oriented towards one entity (xi) involved in one of the subevents
that form part of the state of affairs depicted by the base word. The
variable LT (‘lexical template’) indicates that the event in which this entity
participates can be of any kind, a state (pred′), an activity (do′), or any
other logical structure.
The following representation is to be understood as a subspecification of
the actor template, and corresponds to the traditionally labelled ‘agent
nominals’, which in RRG terms should be described as ‘effector nominaliza-
tions’ (Van Valin and Wilkins 1996).
(15) φiN: [do′ (xi, [φBASE])], x ¼ actor. E.g. driver, runner, smoker.
This structure expresses the semantic content of the most prototypical nom-
inalizations within the class: the derived words corresponding to this con-
struction describe the effector involved in the event expressed by the
semantics of the base word. There are two co-indexing possibilities expressed
in the above representation depending on whether the formation is deverbal
(φV) or not (φ[-V] ). In the case of deverbal effector nouns, co-indexation is
usually quite straightforward: the verbal bases typically encode an event that
is dynamic, and therefore the meaning of effectorhood derives from the
semantic function of its first argument. This is the case of hunter.
(16) hunter
[X-er] N: [ (xi , [huntv ])], x = actor
The representation in (18) also reveals that there are interesting semantic
differences among the members of this subclass. The affix -ista is more
restrictive than other members among the actor nominalizing affixes in
terms of: (a) the types of bases (it prefers nouns and adjectives, in contradis-
tinction with the more deverbal -or and -nte); and (b) its meaning, as it always
involves a degree of volitionality not necessarily present in the other affixes.
Thus, its LT would be more specific or hyponymic.
(19) [φBASE: N/adj þ istaiN] iN : [DO (xi, [do′ (xi, [e2: LT . . .φBASE: (Qualia: . . .LT (xi…)])])]
(20) i
[ BASE + V ] V: [do′ (x, Ø )] CAUSE [ e2 : (LT : … BASE ... )]
i = denotational co-indexation
On the other hand, deadjectival forms like Spanish suavizar ‘soften’, ensan-
char ‘widen’, solidificar ‘solidify’, etc. select the formal quale of the base and
the meaning of the complex word will be that of a prototypical causative-
resultative verb (‘cause become adj’), as shown in (22) (Cortés-Rodríguez and
Sosa 2012: 28).
COMPLEX WORDN
COREN
CORE CORE
NUC NUCN
book sell er
Figure 8.10 Exocentric analysis of compound bookseller (Martín Arista 2009: 92)
The logical follow-on from this statement is that if a word has a referential
semantic function, its nucleus (or head) should be the element that confers
such a value. We can borrow some criteria from Štekauer (2005b: 225–226) to
select the nucleus element within the LSW of a complex word: (i) Hyponymy:
the complex word is a semantic specification of the head; storyteller, for
example, is a hyponym of teller, which in turn is a hyponym of -er. (ii) Subcat-
egorization: heads impose subcategorization restrictions; -ist selects noun
bases, -en has a strong preference for monosyllabic bases with final plosives,
as in red > redden, short > shorten. (iii) Distributional and categorial equiva-
lence: the head (or nucleus) specifies the lexical class of the derived word.
For example, -er formations are consistently denominal; on the other hand,
locative prefixes (pre-field or sub-way) are not heads as they do not determine
the class form of the derived word.
This last criterion draws a key parallel between the nature of the head or
nucleus of clauses and phrases, on the one hand, and complex words, on the
other. If we take these conditions into account, the layered structure of book-
seller could be as shown in Figure 8.11 (Cortés-Rodríguez and Sosa 2012: 36).
The structure captures the endocentric character of the bound lexeme -er
in the derivation of seller and the recursive compositional process with book.
Figure 8.12 shows the structure of non-endocentric complex words in
COMPLEX WORDN
COREN
NUCLEUSN
WORDN PREDICATEN
COMPLEX WORDN
CORE N COREN
NUCN NUCN
COREV
NUCV
PREDICATEV
book sell er
Figure 8.11 Endocentric analysis of compound bookseller (Cortés-Rodríguez and Sosa 2012: 36)
8.5 Conclusion
COMPLEX WORDN
COREN
COREV COREN
NUCV NUCN
PREDV PREDN
lava platos
Figure 8.12 ‘Acentric’ Spanish compound lavaplatos (‘dishwasher’) (Cortés-Rodríguez and Sosa
2012: 37)
between morphology, on the one hand, and lexical semantics and syntax,
on the other. As a lexicological phenomenon, derivational affixes and
processes form part of the lexicon in RRG and – very much like free
lexemes – are endowed with a semantic representation in the format of a
lexical template. The interaction of affixal and word templates is
accounted for by means of two explanatory devices: co-indexation and
qualia specification. The section closed with a brief overview of how the
semantic structure underlying a derivationally complex word also finds its
syntactic counterpart in the LSW, as described in Section 8.2, with some
slight adjustments to the proposal put forward by Martín Arista (2008,
2009, 2011).
In our opinion, taken together, the proposals outlined in this chapter,
which draw on several contributions both from inside and outside RRG,
constitute an explanatory framework for the study of morphological phe-
nomena within this model.
References
Notes
1 We may even include phonology here, though this remains a very under-
developed area. In Diedrichsen and Nolan (2011) there is an initial contri-
bution in this field within RRG, and Möllemann (2016: 30–34) provides a
brief description of the basic conditions necessary for a morphology–
phonology interface in RRG.
Abbreviations
9.1 Introduction
The syntax of adverbs has triggered much controversy both within and
across theories (see, for example, Alexiadou 1997; Cinque 1999; Engels
2012; Ernst 2002, 2020).* The central concern includes the ability to offer
a principled account of (i) which positions an adverb can occupy within a
syntactic representation and (ii) what determines the ordering of mul-
tiple adverbs in sentence structure. Reflecting on recent developments in
the generative approach to adverbial syntax, Ernst notes the critical
consensus that ‘the semantics of individual adverbs is an important
determinant – perhaps the main determinant of their ordering’ (2014:
108). While Ernst is considering generative syntax, the essence of his
* I am grateful to Robert Van Valin, Jr. and to a reviewer for their valuable comments; they greatly improved the quality of
the paper. I thank James Watters for clarifying points on the Tepehua ideophones, and Janis Nuckolls for confirming the
point on the Quechua data. Thanks also go to Elizabeth Thompson for her editorial suggestions. The remaining
shortcomings are, of course, solely my responsibility.
reflection has close affinity with how Role and Reference Grammar (RRG)
analyses adverbs: it too considers semantics a critical component in
theorizing the syntax of adverbs. This affinity may not be accidental, as
a series of Ernst’s own work (e.g. 2002), characterized by him as a ‘scopal
approach’ (Ernst 2014), is rooted in Jackendoff (1972), and RRG draws on
insights from that same work.
Assuming that adverbs can be defined as ‘modifiers of constituents other
than nouns’ (Schachter 1985: 20), the first part of this chapter offers a
foundational sketch of the RRG approach to adverbs using English data,
drawing on Van Valin (2005) and Van Valin and LaPolla (1997), with a focus
on their subsets, including aspectual adverb (completely), manner adverb
(clumsily), temporal adverb (yesterday), evidential adverb (evidently), and
other clausal adverbs. It is pointed out that adverbs modify distinct layers
of the clause – the nucleus, the core, and the clause – and when the
sentence contains multiple adverbs, their appearance corresponds to the
order of the layers they modify in terms of the distance from the verb. The
second part of the chapter moves to an exemplification of how the RRG
approach can be used to analyse the ideophonic adverb, which is typically
a type of manner adverb (Schachter 1985: 21).1 While the category status of
ideophones remains disputed,2 it has frequently been noted that some
function as adverbs, appearing in the same syntactic position as adverbs
of prosaic words (Beck 2008: 41; Bobuafor 2013: 351; Ibarretxe-Antuñano
2006: 19; Nuckolls 1996: 72, among others). For instance, observing an
example like (1), wherein the ideophone expresses a manner of running,
Schaefer notes: ‘Emai IA [i.e. ideophonic adverb] forms assume the canon-
ical syntactic position of adverbs. They occupy [the] postverbal position’
(2001: 341).
SENTENCE
PERIPHERY CLAUSE
PERIPHERY CORE
RP MP MP NUC PERIPHERY PP
COREP
COREM COREM PRED RP MP
NUCP RP
NUCM NUCM COREM CORER
PRED
ADV ADV V NUCM PERIPHERYR NUCR
ADV P MP N
Leslie has evidently been slowly immersing herself completely in the new language.
V N
IF CLAUSE
SENTENCE
Figure 9.1 Example of LSC (modified from Figure 1.13, Van Valin 2005: 22)
evidently modifies the clause, slowly modifies the core, and completely modi-
fies the nucleus. A brief note on the term ‘periphery’ is in order, as it is
sometimes confused with a concept in prototype theory (e.g. Lakoff 1987),
whereby membership in a category is characterized as ‘central’ (i.e. proto-
type) vs. ‘peripheral’ (non-prototypical), or given a configurational region
of ‘central’ (i.e. core) vs. ‘peripheral’ (i.e. margin). As Figure 9.1 makes
clear, in RRG, ‘periphery’ means neither a ‘non-prototypical member’ nor
a ‘margin’ but refers to a structural unit containing a modifier such as an
adverb (Van Valin 2005: 4).
The bottom part of the representation in Figure 9.1 is called the ‘operator
projection’. Operators refer to ‘grammatical categories . . . [that] modify the
clause and its parts’ (Van Valin and LaPolla 1997: 40). They are posited to
have scope over a specific layer of the clause:
The nuclear operators have scope over the nucleus; they modify the
action, event or state itself without reference to the participants. Core
operators modify the relation between a core argument, normally the
actor, and the action . . . Clausal operators, as the name implies, modify
the clause as a whole.
(Van Valin 2005: 8–9)
Again, the arrows in the figure indicate the modificational relations. Inci-
dentally, Figure 9.1 contains only three grammatical categories (ASP, TNS, IF
(i.e. aspect, tense, illocutionary force)), but RRG posits nine operator categor-
ies: (i) aspect, (ii) negation, (iii) directional, (iv) event quantification,
(v) modality, (vi) status,3 (vii) tense, (viii) evidentials, and (ix) illocutionary
force (Van Valin 2005: 12). Each modifies a designated layer (or layers) of the
clause, as represented in Figure 9.2: for instance, aspect is a nuclear oper-
ator. Some of the operator categories become relevant when we discuss the
syntactic characteristics of ideophones.
NUC Aspect
Negation
Directionals
CORE Directionals
Event quantification
Modality
Negation
CLAUSE Status
Tense
Evidentials
Illocutionary Force
SENTENCE
Figure 9.2 Operator projection (adapted from Van Valin 2005: 12)
First, consider the case of the core adverb occurring preverbally. (4) shows
the predicted order of three types of adverbs, and (5) shows different pat-
terns of a sentence containing the three adverbs in different positions.
(4) Order of adverbs when the core adverb precedes the verb:
clausal > core > nuclear
e.g. evidently [evidential: clausal] > slowly [pace: core] > completely
[aspectual: nuclear]
In (5a), the three adverbs appear preverbally, whereas in (5b), the clausal
adverb evidently and the core adverb slowly occur preverbally, but the nuclear
adverb completely is postverbal. The adverbs in (5a)–(5b) observe the predicted
order, rendering the sentences grammatical. In contrast, (5c)–(5g) are judged
ungrammatical, as the adverbs disobey the predicted order given in (4): for
example, in (5c), when the nuclear adverb completely precedes the core
adverb slowly, the sentence is infelicitous; in (5d), when the core adverb
slowly precedes the clausal adverb evidently, the sentence is similarly
infelicitous.
A similar situation arises when the core adverb occurs postverbally. (6)
shows the predicted order, and (7) gives examples with adverbs occurring in
different positions.
(6) Order of adverbs when the core adverb follows the verb:
nuclear < core < clausal
e.g. completely [aspectual: nuclear] < slowly [pace: core] < evidently
[evidential: clausal]
(7) a. Leslie has been immersing herself completely in the new language
slowly evidently.
b. Leslie has been completely immersing herself slowly in the new
language evidently.
c. * Leslie has been immersing herself slowly in the new language
completely evidently.
Among the different patterns, (7a) and (7b) are felicitous, wherein the three
adverbs appear as nuclear, core, and clausal adverbs, in that order, toward the
end of the sentence (although the nuclear adverb appears before the verb in
(7b)). In contrast, the sentences in (7c)–(7g) are infelicitous; they do not observe
the order in (6). For instance, in (7c), the nuclear adverb completely occurs
between the core adverb slowly and the clausal adverb evidently, and in (7d)
the core adverb slowly appears after the clausal adverb evidently.
These examples show that the ordering constraints work in both direc-
tions, substantiating the argument that ‘the scope constraints require that
the nuclear adverb be closer to the verb than the core adverb, and likewise
for the core adverb with respect to the clausal adverb’ (Van Valin 2005: 21).
These relationships are readily depicted in the LSC. Figure 9.3 shows the
SENTENCE
CLAUSE PERIPHERY
CORE PERIPHERY
MP
RP NUC PERIPHERY PP MP
RP
COREP COREM
MP COREM
NUCP RP
PRED NUCM
COREM CORER NUCM
PRED
PERIPHERYR NUCR
V NUCM ADV
ADV
ADV P MP N
Leslie has been immersing herself completely in the new language slowly evidently.
V N
NUC ASP NUCR
ASP NUC CORER
CORE
DEF RP
TNS CLAUSE
IF CLAUSE
SENTENCE
Figure 9.3 Structure of (7a) (modified from Figure 1.14, Van Valin 2005: 22)
constituent and operator projection of (7a) (see Figure 9.1 for the syntactic
representation of (5b)).
Although the examples discussed to this point are English, the fundamen-
tal point about adverb order is expected to be cross-linguistically valid. That
is, regardless of which language is under investigation, the relative order of
adverbs should observe the order of the layers they modify: for example,
adverbs modifying the entire clause should appear away from the predicate,
and adverbs modifying the predicate should appear in closer proximity to
the predicate.
It is worth pointing out that temporal adverbs such as tomorrow (in their
default position) are core modifiers, not clausal modifiers, although the
meaning of temporal adverbs is related to tense, and tense is a clausal
operator (Van Valin 2005: 19; Van Valin and LaPolla 1997: 32, 426–428). This
point is elaborated by Bohnemeyer and Van Valin (2017), who note that
‘verbal cores are inherently constituents that describe (sub)events and that
each core . . . has its own time-positional modifier’ (2017: 159). For instance,
in (8), on Monday modifies the event of persuading, whereas on Friday modi-
fies the event of visiting.
modify the same layer of the clause. In fact, when discussing examples like
(10), Van Valin (p.c.) suggests semantic-based constraints may be at work;
more specifically, if three kinds of core adverbs co-occur (as in (10): a pace
adverb quickly, a manner adverb carefully, and a temporal adverb yesterday),
ordering constraints appear.
First, with respect to position, the pace adverb and the manner adverb seem
to be used interchangeably ((10a) vs. (10c)). Second, the pace adverb and the
manner adverb can be conjoined but neither can be conjoined with the
temporal adverb ((10f) vs. (10g/10h)). Third, in the postverbal position, the
temporal adverb should follow the manner adverb ((10c) vs. (10d)), and
preverbally, the temporal adverb should precede the manner adverb (10j).
The differences in acceptability in (10) strongly suggest some semantic
layering, at least within the core; however, a closer investigation is called
for, with more data.
Temporal adverbs like yesterday in (11a) take the entire logical structure as
their argument (Van Valin 2005: 49). Manner adverbs like elegantly in (11b)
take the activity component as their argument (indicating the elegant
manner in which Pat executed the action of closing), whereas the pace
adverb slowly in (11b) takes the accomplishment component as its argument
(indicating the slowness of the door’s closing event). Finally, in (11c), the
aspectual adverb completely takes the basic state component melted′ (x) as
its argument.
It is left for future work to evaluate whether this uniform treatment of
adverbs, that is, positing a one-place predicate as their semantic representa-
tion, can fully account for the characteristics of adverbs across the board,
either within a given language or cross-linguistically.
Aspect is argued to be part of the lexical meaning of the first two forms
(Akita 2009; Toratani 2005). Singletons typically express an event/state
lasting a short time, such as a semelfactive event, that is, a single-staged
event with no outcome (Smith 1997: 29) (e.g. kon ‘sound of a knock’), an
On the one hand, singletons are obligatorily marked by to, regardless of how
many times the form is instantiated: once (13a), twice (13b), and so forth. On
the other hand, reduplicated forms (13c) and ri-suffixed forms (13d) can be
marked by to or Ø, yielding no truth-conditional differences between the
alternative markings (Hamano 1998, among others; see Akita and Usuki
2016 for a discussion of ‘optionality’ of to-marking). As (13) suggests, the
most frequently occupied position of mimetics of any morphological shape
is immediately preverbal (cf. Toratani 2017: 36–41). However, because Japan-
ese, an OV language, has a flexible word order, clause-internal phrases,
including mimetics, can scramble rather freely as long as the verb comes
at the end of the clause.
(14a) represents the base sentence with no mimetic. The noun hito ‘person/
people’ has an ambiguous reading, in that it can be singular or plural, as nouns
are not obligatorily marked for number in Japanese; yet the default interpret-
ation of the sentence is a single event of coming in. In contrast, (14b)’s mimetic
zorozoro implies multiple events of coming in, thereby depicting the motion of
a crowd of people lining up and moving forward. Therefore, this mimetic is
analysed as a core modifier, as it expresses event quantification.
Other languages have ideophones lexically encoded with the notion of
mass, as exemplified below.
ˇ
(15) a. Chichewa (Bantu; Kulemeka 1993: 224)
u:nji: ‘gather in a mass’
b. Luwo (Western Nilotic, South Sudan; Storch 2014: 44)
m ̀ rm ̀ r
c c ‘a lot of people/cattle moving together’
c. Siwu (Ghana-Togo Mountain; Dingemanse 2011: 48)
ɣèèè ‘animals swarming in great numbers’
According to Nuckolls, ‘dzawn describes the way flies that had gathered on a
molting snake rose up in a swarm when a person approached’ (1996: 149).
The fact that putan ‘fly’, glossed in the singular, is interpreted as plural in
(16) suggests the ideophone plays a role in assigning the mass interpretation.
If this is the case, the ideophone in (16) is a core modifier.
The second grammatical concept applicable to ideophones is aspect. As
noted in (2), aspectual adverbs are assumed to be nuclear modifiers (Van
Valin 2005: 49). Examples are continuously and completely. The assumption is
that ‘aspectual adverbs modify the basic state or activity predicate’ (Van
Valin 2005: 49): e.g. melt completely: BECOME complete′ (melted′ (x)). If ideo-
phonic adverbs can express a concept similar to completely and continuously
and modify the basic state or activity predicate, it seems reasonable to posit
them as nuclear modifiers. Some languages seem to have them, or at least
the English glosses suggest this possibility (e.g. wic ‘eat all up’ (Luwo, Storch
ˇ
2014: 47), psí:tí: ‘completely finished’ (Chichewa, Kulemeka 1993: 252), tdip
‘manner of covering completely’ (Didinga, a Southwest Surmic language,
Sudan, de Jong 2001: 133)). In Japanese, no mimetics precisely express the
meaning of completely and continuously. However, as discussed in Section
9.3.3.2, some mimetics affect the interpretation of the lexical aspect of the
verb when they occur adjacent to the verb and, as such, seem to qualify as
possible nuclear modifiers.
9.3.2.2 Hyponymy
The second semantic criterion to determine which layer of the clause ideo-
phones modify is hyponymy. It has long been noted that some Japanese
mimetics co-occur with a limited set of verbs (Hirose 1981). For instance,
tekuteku ‘plodding’ or yotiyoti ‘toddling’ typically co-occur with aruku- ‘walk’.
In such cases, the mimetic–verb relationship can be characterized in terms of
hyponymy (Toratani 2007: 325–327), where the mimetic is considered the
hyponym and the verb is the hyperonym of the mimetic, as tekuteku ‘plodding’
and yotiyoti ‘toddling’ both express a kind of a walking event.6
In other languages as well, some ideophones collocate or co-occur with a
limited number of verbs/adjectives (Childs 1994: 188; Creissels 2001: 78). For
instance, in Didinga (a Southwest Surmic language in Sudan), the ideophone
ɪðaatʃ ɪðaatʃ ‘manner of swallowing easily’ co-occurs with kú ‘swallow’ (de
Jong 2001: 136), while in Emai (Nigeria’s Edoid group), the ideophone
ghóighói ‘glistering-ly’ co-occurs with jín ‘shine’ (Schaefer 2001: 351), with
the former apparently the hyponym of the latter in each pair.
The hyponymy relationship highlights various ways in which ideophones
and core adverbs modify the predicate’s event. Regular core adverbs can co-
occur with a wide variety of predicates (clumsily walk/dance/write/sit etc.), as
they refer to a general property common to a group of predicates, such as
dynamism, conveying how the event participant performs the action with-
out changing the meaning of the action itself; meanwhile, when ideophones
co-occur with a limited set of predicates, they change the reading of the
action from a more or less neutral manner (e.g. ‘walking’) to a much more
detailed one (e.g. ‘toddling’). To interpret this difference in terms of the layer
of the clause the adverbs modify, we could say the semantic function of
ideophones closely parallels that of nuclear adverbs, in that their functions
are in line with the nuclear operators’ function of ‘modify[ing] the action,
event or state itself without reference to the participants’ (Van Valin 2005:
8–9). This parallelism suggests ideophones that co-occur with the verbs/
adjectives of their hyperonymous category are nuclear adverbs, unless they
refer to multiple event participants, in which case, the ideophones are core
modifiers.7
The preceding discussion may give the impression that all ideophones are
hyponyms (cf. Watson 2001: 393). This is not the case. To draw again on
Japanese data, while some mimetics are clearly hyponyms (candidates for
nuclear adverbs) as exemplified in (17), others are non-hyponyms (candidates
for core adverbs).8
These observations imply that ideophones are not clausal modifiers. This
possibility can be confirmed with Japanese data. As noted in Section 9.2.1.2,
typical clausal adverbs express the speaker’s evaluation or judgement of a
propositional content (e.g. unfortunately, probably). Mimetics do not seem to
convey these concepts. Furthermore, like mimetics, the large majority of ideo-
phones express sound, manner, and the state of an entity, detailing the actions
and states expressed by the clause-mate verbs/adjectives, but they do not seem
to convey any concepts paralleling those conveyed by clausal adverbs. This
suggests that ideophones are not clausal adverbs cross-linguistically as far as
the meaning is concerned, but this suggestion requires validation.9
Example (20) contains the mimetic batan ‘(sound of ) a bang’ in three differ-
ent preverbal positions. As the identical gloss (with a bang) indicates, it yields
only one reading, that of manner, irrespective of where the mimetic occurs
within the sentence.
This unambiguity of reading is supported by the paraphrasability of the
meaning of the mimetic into a manner reading (21a) but not into an
evaluative (i.e. clausal adverb) reading (21b).
In other words, the mimetic remains a core modifier and cannot be clausal,
even if it changes syntactic position. This leads to the following two predic-
tions. First, the mimetic should be able to co-occur with a clausal adverb:
more specifically, because the mimetic and the clausal adverb modify dis-
tinct layers of the clause, they will not cause a semantic clash, unlike two
clausal adverbs as in evidently and probably in *Evidently, John probably left
(Jackendoff 1972: 87). Second, the mimetic should observe the order con-
straint with respect to the clausal adverb. That is, the clausal adverb should
occur first, followed by the mimetic. (22) and (23) illustrate these points (the
mimetic gatyan ‘sound of a clank’ is a core modifier as it is not a hyponym of
otosu ‘drop’).
The first point is illustrated in (22) by the ability of the mimetic gatyan
‘sound of a clank’ to co-occur with the clausal adverb bukiyooni-mo ‘clum-
sily’.10 The second point (word order) is also illustrated in (22) by the
expected word order: the clausal adverb bukiyooni-mo ‘clumsily’ precedes
the mimetic (core/nuclear adverb). This is further supported in (23) by the
mimetic’s inability to cross the clausal adverb.
To reiterate, unlike English manner adverbs like clumsily, mimetic adverbs
do not yield a clausal reading, even if they change syntactic position.
Though verification is necessary, this seems to apply to ideophones
across languages.
which the to-marked version of the mimetic moved out of the immediately
preverbal position.
(26) The ideophone that specifies the aspect of the predicate is a nuclear adverb;
all other ideophones are core adverbs.
SENTENCE
CLAUSE
PERIPHERY CORE
RP PP PERIPHERY NUC
MP PRED
MI-ADV V
Mizu-o gohunkan gokugoku non-da.
water-ACC for.5.min MIMETIC drink-PST
‘The child gulped water for five minutes.’
Figure 9.4 The mimetic as a nuclear adverb (adapted from Toratani 2007: 333)
SENTENCE
CLAUSE
PERIPHERY CORE
NUC
RP MP RP PP
PRED
MI-ADV V
Kodomo-ga gokugoku-to mizu-o gohunkan/gohun-de non-da.
child-NOM MIMETIC-P water-ACC for/in 5 minutes drink-PST
‘The child, in a gulping manner, drank the water in/for five minutes.’
Figure 9.5 The mimetic as a core adverb (modified from Toratani 2007: 334)
As Figures 9.4 and 9.5 show, when they are at clause-internal positions,
mimetics are contained in the periphery and modify a specific layer of
the clause.
This chapter outlines the RRG approach to adverbs, paying particular atten-
tion to ideophonic adverbs. It has introduced relevant RRG assumptions on
adverbs: (i) that adverbs contained in the periphery and represented in the
‘constituent projection’ of the layered structure of the clause may modify all
three layers of the clause, that is, the nucleus, the core, and the clause;
(ii) that the order of adverbs modifying a distinct layer of the clause observes
the order of the layers of the clause with respect to the position of the verb,
as the nuclear adverb is constrained to occur closer to the verb than the core
adverb, which, in turn, is constrained to occur closer to the verb than the
clausal adverb.
The RRG approach to adverbs remains rather preliminary (cf. Cortés-
Rodríguez and Rodríguez-Juárez 2019), dealing only with basic phenomena.
More analyses are certainly required, for instance, to identify the ordering
constraints among the adverbs modifying the same layer, considering their
motivations, or to substantiate how the logical structure of ideophonic
adverbs can be represented (for example, does this call for a one-place
predicate just like adverbs of prosaic words?) to capture the unique seman-
tics of ideophones.
References
Bohnemeyer, Jürgen and Robert D. Van Valin, Jr. 2017. The macro-event
property and the layered structure of the clause. Studies in Language 41
(1): 142–297.
Childs, Tucker G. 1994. African ideophones. In Leanne Hinton, Johanna
Nichols and John J. Ohala (eds.), Sound Symbolism, 178–204. Cambridge:
Cambridge University Press.
Cinque, Guglielmo. 1999. Adverbs and Functional Heads: A Cross-Linguistic Per-
spective. Oxford: Oxford University Press.
Cortés-Rodríguez, Francisco, J. and Carolina Rodríguez-Juárez. 2019. The
syntactic parsing of ASD-STE100 adverbials in ARTEMIS. Revista De Lingüís-
tica y Lenguas Aplicadas 14: 59–79.
Creissels, Denis. 2001. Setsuwana ideophones as uninflected predicative
lexemes. In E. F. K. Voeltz and C. Kilian-Hatz (eds.), Ideophones, 76–85.
Amsterdam: John Benjamins.
Croft, William and D. Alan Cruse. 2004. Cognitive Linguistics. Cambridge:
Cambridge University Press.
De Jong, Nicky. 2001. The ideophone in Didinga. In E. F. K. Voeltz and C.
Kilian-Hatz (eds.), Ideophones, 121–138. Amsterdam: John Benjamins.
Dingemanse, Mark. 2011. The Meaning and Use of Ideophones in Siwu. Unpub-
lished PhD dissertation, Radboud University Nijmegen.
Dingemanse, Mark. 2017. Expressiveness and system integration: On the
typology of ideophones, with special reference to Siwu, STUF – Language
Typology and Universals 70(2): 363–385.
Dingemanse, Mark. 2019. ‘Ideophone’ as a comparative concept. In K. Akita
and P. Pardeshi (eds.), Ideophones, Mimetics, and Expressives, 13–33. Amster-
dam: John Benjamins.
Doke, Clement Martyn. 1935. Bantu Linguistic Terminology. London: Longmans,
Green, and Co.
Engels, Eva. 2012. Optimizing Adverb Positions. Amsterdam: John Benjamins.
Ernst, Thomas. 2002. The Syntax of Adjuncts. Cambridge: Cambridge University
Press.
Ernst, Thomas. 2014. The syntax of adverbs. In A. Carnie, D. Siddiqi and Y.
Sato (eds.), The Routledge Handbook of Syntax, 108–130. New York: Routledge.
Ernst, Thomas. 2020. The syntax of adverbials. Annual Review of Linguistics 6(1):
89–109.
Hamano, Shoko. 1998. The Sound-Symbolic System of Japanese. Stanford, CA: CSLI
Publications.
Hirose, Masayoshi. 1981. Japanese and English Contrastive Lexicography: the Role
of Japanese ‘Mimetic Adverbs’. Unpublished PhD dissertation, University of
California, Berkeley.
Ibarretxe-Antuñano, Iraide. 2006. Sound Symbolism and Motion in Basque.
Munich: Lincom Europa.
Jackendoff, Ray S. 1972. Semantic Interpretation in Generative Grammar. Cam-
bridge, MA: MIT Press.
Notes
10.1 Introduction
This chapter has a twofold aim: to discuss the RRG treatment of adpositional
phrase (AP) types and adpositional assignment (Foley and Van Valin 1984;
Van Valin 1991, 1993, 2005; Van Valin and LaPolla 1997; Jolly 1993) and to
propose a new typology of adpositional phrase types, following Ibáñez Cerda
(2009, 2011). Although we only discuss data from English and Spanish, the
insights and analyses presented here can be broadly applied to other
languages.
The RRG treatment of adpositional systems relies on three principal
assumptions. First, the function of adpositions is comparable to that of
morphological case, in that adpositional assignment may depend on the
semantic role of the argument. In this sense, adpositions can be considered
to be analytic case forms (see Chapter 7). Second, RRG assumes that the
assignment of adpositions is not idiosyncratic and need not be postulated as
part of the lexical entries of verbs. Instead, adpositional assignment follows
systematic rules, which are applied in the linking and, rather than being
based on grammatical relations or phrase structure positions, depend on the
semantic content of the logical structure (LS) of the predicate and on the
semantic/syntactic distinction between direct and oblique core arguments.
The third assumption is the distinction, based on Bresnan (1982), between
predicative and non-predicative adpositions. Non-predicative adpositions
mark verbal arguments: they do not license these arguments and they do
not add any substantive semantic information to the clause. They are a
function of the semantics of the predicate and, thus, they are free-
morphemic case markers assigned by the predicate (e.g. to in Pat gave a book
to Peter). Contrastingly, predicative adpositions are predicates, in that they
contribute substantive semantic information to the clause in which they
occur, both in terms of their own meaning and of the meaning of the
argument that they license. A typical example of a predicative adposition
As is stated in Van Valin (2005: 4), RRG bases its analysis of clause structure
on two universal semantic distinctions: the one between the predicate and
non-predicating elements and, on the other hand, among the non-
predicating elements, the distinction between arguments and non-argu-
ments (see Figure 10.1).
The distinction between arguments and non-arguments is based on the
idea that the states of affairs represented by the predicates inherently
determine the number and type of participants involved in them. The
nature of the situation conditions the presence of the participants that are
predicate
Non-arguments
arguments
Figure 10.1 Universal oppositions underlying clause structure (Van Valin 2005: 4)
CLAUSE
CORE PERIPHERY
NUCLEUS
Figure 10.2 Components of the layered structure of the clause (Van Valin 2005: 4)
needed in order to make possible the state of affairs. In this view, the
participants inherently required by a predicate are the arguments, while
those that are not are the non-arguments or adjuncts.
Based on these semantic oppositions, RRG proposes the layered structure
of the clause (LSC), which consists of the nucleus, the core and the periphery,
as represented in Figure 10.2.
Similarly to Lexical Functional Grammar (Bresnan 2001) and Functional
Grammar (Dik 1989), RRG assumes that syntax and semantics work in
parallel. There are no derivations, or any deep levels of representation, but
just one single level with two parallel representations.1 There are natural
correspondences between syntax and semantics. To begin with, the semantic
predicate corresponds to the nucleus of the clause in syntax. Following an
iconicity principle (Haiman 1980), arguments, along with the nucleus,
belong to the core, whereas the non-arguments are peripheral elements of
the clause. Depending on their morphosyntactic properties, arguments can
be direct or oblique. The direct core arguments are marked by direct mor-
phological case: nominative and accusative in nominative-accusative
systems, and absolutive and ergative in absolutive-ergative systems. Periph-
eral non-arguments are adjuncts. Against this backdrop, a clause like John
gave a book to Mary in the library can be represented as in Figure 10.3.
CLAUSE
CORE PERIPHERY
John gave a book to Mary in the library
NU
NUCLEUS
Figure 10.3 Layered structure of John gave a book to Mary in the library
Using the notions just outlined, we can say that John and a book are direct
core arguments, while to Mary is an oblique core argument. The three of
them are participants semantically required by the verb to give. As for the
library, it is an adjunct. Table 10.1 summarizes the relation between the
units of the syntactic and semantic representations.
A third type of clausal component, originally proposed by Jolly (1993), is
that of argument-adjuncts in the core. This complement type corresponds to
APs that code a verbal argument but, at the same time, are introduced by a
predicative preposition.
Table 10.1 Semantic units underlying the syntactic units of the layered structure of
the clause (Van Valin 2005: 5)
Predicate Nucleus
Argument in semantic
representation of predicate Core argument
Non-argument Periphery
Predicate þ Arguments Core
Predicate þ Arguments þ
Non-arguments Clause (¼ Core þ Periphery)
Although most verbs follow these defaults assignments, there are excep-
tions (e.g. intransitive two-place predicates such as locative or psych verbs).
These must be specified in the lexical entries of the relevant verbs in terms
of a simple feature [MR α], with values [MR 0], [MR 1] and [MR 2].
Finally, the relation between macroroles and logical structure argument
positions is captured in the Actor-Undergoer Hierarchy (AUH) in Figure 10.4
(Van Valin 2005: 61).
ACTOR UNDERGOER
Given the logical structure of a transitive verb, the leftmost argument will
by default be the actor and the rightmost argument will be the undergoer.
Marked assignments to undergoer are possible, typically with three-place
predicates, where there can be two arguments competing for a macrorole
function, as in (4a–c).
Example (4c) illustrates the dative shift alternation, whereby the first argu-
ment of the two-place state predicate, not the second, is assigned the
macrorole undergoer.
forms, although this type of case marking goes beyond the scope of
this chapter.
In RRG, the notion of case marking is extended to cover the function of the
cross-referencing pronominal affixes which appear on the verbal base in
head-marking languages (Nichols 1986, Van Valin 2005: 16–17). These affixes
perform a twofold function: on the one hand, they indicate the relation
between the arguments and the verb, and on the other hand, they are
arguments in syntax. Only the grammatical or direct cases are marked on
the head verb; other semantic relations, if marked, will be marked through
adpositions or morphological case on the dependents.
(5) Kim put the book {in / on / next to / behind / on top of / under} the box.
PP
COREP
PP
NUCP RP
PRED
P RP
P
In this section, we discuss the RRG treatment of the three types of adposi-
tional phrase introduced above: predicative (10.4.1), non-predicative (10.4.2)
and argument-adjunct (10.4.3), drawing on Jolly (1993), Van Valin and
LaPolla (1997) and Van Valin (2005).
In (6a), Chris’s running takes place in the park, and therefore the logical
structure of the predicative preposition in is the highest predicate in the
logical structure; it takes the park and the logical structure for run as its two
arguments. This contrasts with the logical structure of a clause like (6c) Chris
ran to the park, where the PP expresses the location of the referent of Chris,
not the location of the event of running, and in this example to the park is an
argument-adjunct PP (see Section 10.4.2).
As pointed out in Ibáñez Cerda (2009), predicative or setting PPs are like
adverbs, in that they can modify different parts of a verbal LS, and not
necessarily all of it, as shown in (7) for Spanish.
Here the PP sobre la tabla de madera ‘on the wooden board’ only refers to the
place where the change of state takes place; more precisely, the wooden
board is the place where the event of ‘dividing the cake into pieces’ happens.
Certainly, Juan is doing something that involves the wooden board, but,
clearly, he is not located on the board. Consequently, the semantic scope of
the PP is not over the core but only part of it. In this way, we can consider
PPs like the one in (7) as partial modifiers of the core. A possible representa-
tion for (7) is thus shown in (8).
(8) [do′ (Juan, Ø)] CAUSE [be-on′ (tabla, [BECOME cut′ (pastel)])]2
In contrast to the LS in (6b), here the adverbial predicate be on′ only has
scope over the subevent of change of state, leaving the activity subevent out.
Apart from the case of (7), the fact that adjunct PPs can modify structural
elements other than the whole core can be seen in a sentence with an added
instrumental complement (9).
Not only is the change of state under the scope of the locative PP, but
part of the action performed by the effector falls inside its scope: John’s
knife manipulation and, more clearly, the contact of this instrument
with the cake, takes place on the wooden board. This can be represented
as follows.
(10) [do′ (Juan, use′ [Juan, cuchillo])] CAUSE [[be-on′ (tabla, [do′ (cuchillo, [cut′
(cuchillo, pastel)])])] CAUSE [BECOME cut′ (pastel)]]
Here, only one of the activity subevents, the one that has the instrument as
an effector, is under the scope of be-on′, while the more external one, the
one where Juan is the effector, falls out of the scope of the preposition.
In sum, following Ibáñez Cerda (2009), we have argued that the PPs in (7)
and (9) behave differently from the one in (6a): the whole core of the clause is
one of the arguments of the PP in (6a), but this is not the case with (7) and (9).
Nevertheless, none of these PPs codify verbal arguments, they are all headed
by a predicative preposition, and they are all modifiers in the periphery. In
this sense, they represent different cases of adjunct complements intro-
duced by predicative PPs.
In each of the LS segments in (15a–c), from marks the first argument of the
two-place state predicate, which is a non-macrorole core argument. The
difference between these segments and those in (13a–c) is the presence of
NOT. This difference in content between to and from was first proposed by
Gruber (1965). Again, like to, from does not mark a single thematic relation,
but rather it is assigned in a particular LS context – BECOME/INGR NOT pred′
(y, z), and the rule (11b) rightly predicts its appearance.
The rule (11c) for preposition with applies in the single-prime examples in
(16a′, b′), where the theme argument z is not selected as undergoer and is
instead marked by with. In contrast, the non-prime examples (16a, b) obey
rule (11a), which requires the marking with to.
Thus, with is assigned in contexts where two arguments can be selected for
the undergoer status, but the marked option – in terms of (11a) – is the one
that prevails.
There are two further contexts where with is assigned: in the marking of
an instrument (17a) and in a comitative PP (17b).
As can be seen from the prime examples in (22), de ‘of’, is the preposition
that, in effect, can appear in all contexts of non-default projection of a
theme, when the competition for the undergoer function is between asym-
metrical arguments, that is, two arguments which do not have the same
status in LS. The domain of con ‘with’ is instead much narrower.
One way of analysing these facts is to consider de to be the basic prepos-
ition for marking non-prototypical argument projections in Spanish. In this
analysis, Spanish de – including the one with genitive function – is an empty
preposition which marks asymmetric relations in non-default cases: (i) lexic-
ally asymmetrical relations between the two arguments of M-intransitive
verbs with the U¼x specification; (ii) non-default coding of asymmetrical
arguments; and (iii) the asymmetrical relations between a noun and its
modifiers. We can thus posit the rule in (23) for the assignment of the de
preposition to verbal arguments in Spanish.
The dative preposition a can still be the default preposition for non-
macrorole arguments, whereas con ‘with’ is the preposition for: (i) the non-
actor co-agent in comitative constructions (e.g. Lola fue al cine con Domingo
‘Lola went to the movies with Domingo’); (ii) the non-actor instrument of cut
and break verbs; and (c) the non-default coding of arguments with locative
and spray verbs.3
Note that the instrumental case rule in (11c) can still partially apply for
the assignment of the preposition con in Spanish.
(24) Assign con to non-MR b argument if, given two arguments, a and b, in a
logical structure with (i) both as possible candidates for a particular
macrorole and (ii) a being equal or higher (to the left of b) on the AUH, b is
not selected as that macrorole.
This rule effectively covers the cases where con competes with de for the
marking of theme arguments not selected as undergoer (locative and spray
verbs) and cases where de does not appear: for example, those where there
are two arguments competing for the actor macrorole (comitative and
instrument cases). De, on the other hand, appears in marked cases where a
theme argument is not selected as undergoer. Thus, the rule in (23) must be
revised as in (25).
Although both (24) and (25) apply with locative and spray verbs, elsewhere
they have their own niche of functionality.
10.4.3 Argument-Adjuncts
Currently, two kinds of argument-adjunct adpositional phrases are recog-
nized in RRG: (a) those that are headed by an adposition which has semantic
content but nonetheless marks a verbal argument, and (b) those phrases
headed by a predicative adposition which introduces a participant not
licensed by the verb (i.e. an adjunct) but does not take the whole LS of the
verb as an argument, sharing its argument with the verbal LS. Both types are
core phrases in syntactic terms.
A typical example of the first type of argument-adjunct is the variable
preposition which codes the goal argument of motion and change-of-place
verbs. Spanish examples are provided in (26) (see also the English example
in (5)).
Ibáñez Cerda (2009) points out that these PPs have different status
depending on the preposition that introduces them. In Spanish, if a goal
argument is coded by a PP introduced by the preposition a ‘to’, with intransi-
tive motion verbs, and by en ‘in/on’, with change-of-place verbs, it should be
analysed as an oblique core argument. Indeed, these prepositions are canon-
ical with the said verb classes. A corpus-based study (Ibáñez Cerda 2005)
shows that the goal PP of intransitive motion verbs was coded with a in over
90 per cent of cases. Similarly, the PP of the change-of-place verbs strongly
tended to be introduced by en. These prepositions are thus assigned system-
atically as the unmarked prepositions in the following contexts: a appears
when the LS of a predicate has a BECOME pred′ (z, y) segment, which is the
case of motion verbs, and en shows up in the structural environment
characterized by the presence of INGR pred′ (z, y), which characterizes
‘putting’ verbs.
By contrast, when the goal arguments are introduced by other prepos-
itions that have more semantic content, such as dentro ‘inside’, hacia
‘towards’, etc., they should be considered as argument-adjuncts in the core,
because although these kinds of prepositions are clearly predicative, the
possibility of using them for introducing the goal argument of change-of-
place verbs is not entirely free, as is the case of the locative PPs that do have
the status of adjuncts. Poner ‘to put’, the predicate that serves as nucleus in
the sentences exemplified in (26b), is the hypernym in the domain of
change-of-place verbs. As such, it has a very general and abstract locative
meaning; it does not inherently specify much about its goal, and that is
why it can be used with goals introduced by almost any locative prepos-
ition. This is not the case, however, with other verbs whose inherent
semantics blocks the use of certain prepositions for coding their goals, as
can be seen in (27).
(30) [[do′ (Robin, Ø)] CAUSE [BECOME baked′ (cake)]] PURP [BECOME have′
(Sandy, cake)]
(31) Juani hizo un traje para Pedro j y lo i /??j usó todo un día.
John made a suit for Peter and acc.3sg wore all one day
‘John made a suit for Peter and (he) wore it one complete day.’
Even though the suit was made for Pedro, the most natural reading of this
example is that John was the one who wore it for a day (perhaps to check
how it looked on him before giving it to Pedro).
These facts suggest that, in marked contrast with the goal arguments of
motion and change-of-place verbs, benefactives introduced by para are not
core arguments. This suggestion is further supported by the fact that Span-
ish has an alternative mechanism for making core arguments out of bene-
factives; this is the dative or indirect object construction, where the
beneficiary is introduced by the preposition a, and in which it can be
doubled by the clitic pronoun le. We refer to Ibáñez Cerda (2009) for more
in-depth discussion on the different behaviour of benefactives, on the one
hand, and datives and recipients, on the other.
Another independent criterion to determine that the PPs introduced by
para in Spanish do not belong to the core is that there are no non-predicative
uses of this preposition. The Spanish counterparts of the English verbs to long
and to hope, which appear with non-predicative for, are mostly transitive (e.g.
espero las buenas nuevas ‘I’m waiting/hoping for good news’). When they are
used intransitively, they are coded with the preposition por, and not with
para (e.g. espero por la buena nueva).
In sum, PPs introduced by para in Spanish are always predicative and do
not behave as core arguments: they code a participant that is not part of the
verb semantics. Nevertheless, they are not like the adjunct PPs analysed in
Section 10.4.1, because they share one argument with the LS of the verb and,
although they are introduced by a predicative preposition, they do not take
the whole clause as one of their arguments. Our proposal is that they are
argument-adjuncts in the periphery.
As pointed out by Jolly (1993) and Van Valin and LaPolla (1997), the
preposition for is a good example of an adposition which can have the three
main functions discussed above: it can be a non-predicative case marker in
an oblique argument PP (32a), it can function as an argument-adjunct, as in
(29), which is repeated in (32b) for convenience, and it can appear as a
predicative preposition heading an adjunct phrase (32c).
As a way of unifying these three uses, Jolly (1993) proposes this extended
semantic representation, which is abbreviated as PURP [. . .] in the logical
structure in (30).
(34) [want′ (Robyn, [BECOME have′ (Sandy, cake)])] ^ [[DO (Robyn, [do′ (Robyn, Ø)]
CAUSE [BECOME baked′ (cake)]] CAUSE [BECOME have′ (Sandy, cake)])]
(35) [want′ (Rita, [BECOME entertained′ (the students)])] ^ [[DO (Rita, [do′ (Rita,
[sing′ (Rita)])])] CAUSE [BECOME entertained′ (the students)]]
Of the eight logical possibilities, only the last type, that is, the one defined
by the negative value of all the features, is ruled out by functional prin-
ciples: there is no way, it seems, in which a semantic adjunct, with no core
privileges, can be introduced by a non-predicative preposition. In the
following section, examples from English and Spanish of each of the other
seven types are provided.
(þ) Argument
(þ) Core
() Predicative preposition
These are canonical oblique core arguments in RRG terms: semantic argu-
ments of the predicate in the nucleus of the clause, which are coded as APs
introduced by a non-predicative adposition. This is the case with the recipi-
ent argument of transfer verbs in English (36a) and of the goal of motion
and change-of-place verbs (36b–c), when they are introduced by canonical
prepositions, like a and en in Spanish, respectively, as these prepositions
are predictable from specific positions in the LS of those predicates.
() Argument
() Core
(þ) Predicative preposition
These are canonical peripheral clause participants. They are not semantically
required by the predicate in the nucleus of the clause, and thus they are
adjuncts. They cannot function as controllers, and hence they must be outside
of the core, that is, in the periphery. Finally, their preposition is predicative
and, as such, it licenses the participant as its argument. The most straightfor-
ward examples of this type of PP are the temporal and locative adjuncts that
function as settings of the state of affairs denoted by the predicate.
(þ) Argument
(þ) Core
(þ) Predicative preposition
(þ) Argument
() Controller
() Predicative preposition
agent PP of the passive construction in English, in which the agent loses all
its syntactic privileges and, as a result, it cannot be projected as a core
argument. In fact, there are languages in which only the direct arguments
are in the core; that is, all oblique arguments seem not to have control or
pivot functions or any other type of syntactic privilege (Bickel 2003).
Other possible examples of this type are those APs which code participants
that are semantically required by the verb, for instance arguments, like en
francés in (39a) and con un gesto in (39b), but in an alternative construction
can be projected as subject or as direct object, as can be seen in (39c) and
(39d). This behaviour indicates that they are indeed arguments.
Importantly, when these participants are coded as APs, they are clearly
optional and peripheral, so they have the status of arguments in
the periphery.
(þ) Argument
() Core
(þ) Predicative preposition
() Argument
(þ) Core
(þ) Predicative preposition
This AP type projects an adjunct, that is, a participant that is not required by
the predicate in the nucleus of the clause. Nevertheless, the (þ) core feature
indicates that, contrary to what is expected of adjuncts, this kind of AP
exhibits some important syntactic properties which give it core status. This
is the case with some Spanish manner, temporal and locative PPs, exempli-
fied in (41a–c).
The PPs in (41a–c) are obligatory for the grammaticality of these clauses, as
can be seen from the comparison with their counterparts in (41d–f), which
are pragmatically odd and need some addition to become acceptable (e.g.
Finalmente, el puete fue construido ‘In the end the bridge was built’). Neverthe-
less, the participants projected through them are not semantically required
by their respective nucleus predicates, so they are adjuncts in the core.
() Argument
(þ) Core
() Predicative preposition
As can be seen from the contrast between (42c–d) and (42e–f), the PP of
(42c–d) obligatorily co-occurs with the clitic le, whereas this is not the case
with (42a–b). However, both types of dative PP can control pivots in non-
finite subordinate clauses.
10.6 Conclusion
References
Notes
4 Almost all of the proposed labels for the different PP types have been used
before in RRG studies, but they also arise directly from the combination of
features used in this work. For example, the label Argument-Adjunct in
the Core comes from the fact that the PP has the features (þargument),
(þcore) and (þpredicative preposition). This last feature gives the label the
‘adjunct’ part of the name, as it is a standard characteristic of adjuncts to
be formed with a predicative preposition. Nonetheless, the adjunct part is
attached to the ‘argument’ part of the name, because coding a participant
which is an argument is the main feature of this type of PP. I am now
introducing the label ‘Adjunct-Argument in the Core’ by means of the
same procedure: the combination of features is (argument), (þcore) and
(predicative preposition). That is, the main feature is that the partici-
pant coded in this type of PP is a semantic adjunct, so the first part of the
proposed name is ‘adjunct’. The ‘argument’ part of the name comes from
the fact that it is a standard characteristic of arguments to have non-
predicative prepositions. Finally, the ‘core’ part indicates that the PP has
some syntactic privileges.
11.1 Introduction
* I am grateful to Mitsuaki Shimojo, with whom I discussed several of the issues dealt with in the chapter.
between the notions of topic and focus, which are defined in terms of the
relations established between, and within, pragmatically structured propos-
itions and, on the other hand, the status of discourse referents in the minds
of the speech act participants. A key role in the RRG treatment of infor-
mation structure is also played by the distinction between presupposition
and assertion. The presupposition of an utterance is the information that is
shared by speaker and hearer prior to a sentence being uttered. The asser-
tion is the information that is known to the hearer as a result of the
sentence being uttered. Pragmatic relations and states will be dealt with
in Section 11.2. Then, in Section 11.3, we introduce the positions on the
layered structure of the clause that are motivated in terms of the encoding
of pragmatic relations.
The contrast between presupposition and assertion is reflected in focus
structure (Lambrecht 1994), that is, the conventional associations of focus
meanings with sentence forms. In accordance with the goal of typological
adequacy (Van Valin and LaPolla 1997: 8), these associations are assumed to
vary across languages. Focus structure is discussed in Section 11.4, where we
also introduce the distinction between the potential domain of focus in the
syntax of a given language and the actual domain of focus in an utterance of
that language. We then provide examples of cross-linguistic variation in
focus structure that depends on the variation in the potential domain of
focus as well as on the respective roles of prosody and syntax in the expres-
sion of discourse-related meaning (Section 11.5).
In Section 11.6 we discuss the pervasive role of discourse in linguistic
production and processing, illustrating it with reference to various steps
in the semantics–syntax linking. Lastly, we draw some conclusions
(Section 11.7).1
(2) a. The student I saw outside your door asked me if I knew where you were.
b. That student asked me if I knew where you were.
expectation that the hearer will be able to single out the sole individual that
is being referred to with this expression in the given utterance.
The notion of frame or schema (Fillmore 1982; Lambrecht 1994: 90) can be
relevant to the treatment of a referent as an identifiable one. A frame or
schema is a system of associated concepts that are related in such a way that
to understand one of them one has to understand the whole system. These
systems of concepts justify the treatment of certain referents as identifiable,
because, when one of them is introduced, all the associated ones become
available. For example, if one is familiar with the concept of an academic
department, one will know that departments have heads. Once the notion of
department has been introduced, then the concept of the department’s head
becomes available.
Referents like those in the italicized expressions in (2a–b), which are to
some extent cognitively available with reference to the linguistic context or
(co-)text, are said to be textually accessible (recall that the intended context
of 2a–b is 1a–b). In some cases, these are referents that have been introduced
at an earlier stage in discourse and have not been mentioned for a while.
Other forms of cognitive availability are inferential or situational accessi-
bility, namely the property of an entity or an individual to be individuated
in the physical context, as is the case with (4b) or by means of some relation
with an element in the physical or linguistic context. The latter case is
illustrated in (5), where the referent of the noun phrase his head of department
may not be known to the hearer, but this referent is nonetheless accessible
by virtue of its relation to another referent (the latter is encoded by the
third-person pronoun his).
Accessible referents are not in the current focus of attention prior to the
sentence being uttered (Van Valin and LaPolla 1997: 200). Identifiable refer-
ents that are in the hearer’s long-term memory, but neither in the short-term
memory nor in the current focus of consciousness, are called inactive,
whereas referents that are in the hearer’s current focus of consciousness are
activated or active.5 An active referent is encoded by them in (6).
(6) Remember John and Pete, the identical twins from high school? I saw them
in the lecture today.
Referential
identifiable unidentifiable
Belloro points out that the object doubled by a clitic is not normally indefin-
ite in the dialects that she investigates. This fact would be puzzling, should
the key property of the doubled object be specificity, as was claimed in
earlier literature. Specifics can, in fact, be formally indefinite. While a
semantic analysis in terms of specificity is challenged by the evidence, an
information structure account in terms of identifiability would also seem to
be problematic on empirical adequacy grounds, since it is not the case that
all definite objects are doubled. The failure of some definite objects to be
doubled is captured with reference to the gradience of identifiability (see
Figure 11.1). Belloro (2015: 39) argues that the doubled objects encode
identifiable referents that are not active, but rather accessible. This claim
explains the facts exhaustively, predicting that the construction will not
exhibit indefinites, since indefinites encode unidentifiable referents or spe-
cifics which are unidentifiable from the perspective of the hearer. On the
other hand, it can host generics, in which case the discourse interlocutors
identify a class (Lambrecht 1994: 82, 88). An example of doubling of a generic
object is given here.
believe, however, that the two notions ought to be kept apart, in that only
referential topics are established in previous context. In fact, in textual
sequences, when a new aboutness topic is introduced, a previous aboutness
topic can take the role of a referential topic, which, however, will not play a
role in the topic-comment articulation of the utterance. The aboutness topic
of (12a), the proposal, turns into a referential topic in (12b), where the new
aboutness topic the teaching assistants is introduced.
utterances, but they are so with respect to the whole proposition of which
they are components.8 Accordingly, focus is defined with reference to the
pragmatic relation between two components of a proposition, the one
being presupposed, the other being the new contribution of the assertion,
or the contribution sought with a question. Thus, the notion of focus is
relational, on a par with that of topic. In the last analysis, the notions of
topic and focus are to be defined in terms of the speaker’s assessment of
the pragmatic relations between the components of a proposition in a
given discourse context. These notions will be discussed further in subse-
quent sections, after we have introduced other aspects of the RRG
approach to information structure.
Before we conclude this section, we return to the issue of the alignment of
cognitive states with pragmatic relations. Underlying the correlation
between these two types of pragmatic notion is the following principle:
the more accessible the topical information unit is, the less effort is required
to process an utterance. Accordingly, accessibility strongly aligns with top-
icality. On the basis of this principle, a Topic Acceptability Scale was
proposed by Lambrecht (1994: 165) and a slightly revised version was then
adopted in RRG (see Figure 11.2).
Recall that in the discussion of the cognitive states of discourse referents,
we pointed out that different strategies are adopted for the encoding of
different degrees of accessibility, thus ensuring that the least amount of
processing effort is made by the hearer. Zero marking tends to express
activation, while pronominal marking tends to correspond to activation or
accessibility, and so on. If we combine these coding strategies, ordered in
terms of the cognitive states that they express cross-linguistically, with the
Topic Acceptability Scale, we obtain the hierarchy shown in Figure 11.3,
which expresses the likelihood of coding of topic and focus by means of the
strategies that mark arguments in terms of their relative accessibility.
Figure 11.3 Coding of referents as topic and focus (Van Valin and LaPolla 1997: 205)
The pre-core slot can also, however, host topics in languages with a V2
constraint on word order (Diedrichsen 2008). As for the post-core slot, it
has been identified as the position of secondary foci that non-canonically
occur in postverbal position in Japanese, which is a verb-final language
(Shimojo 1995).
The pre-detached position can iterate, thus allowing the utterance to have
several topics, while the pre- and post-core slots cannot be repeated.
The core-internal positions and the peripheries of the various layers of the
clause can also host constituents with particular discourse roles. For
example, aboutness topics occur in the core-internal immediately pre-
nuclear position in SV languages. However, these positions are not pragmat-
ically motivated. In other words, they are not defined in terms of the
discourse roles that they host or in terms of the constructions in which they
are involved and which have pragmatic conditions on their occurrence. In
Figure 11.4, we indicate with grey shading, and lack thereof, the positions
that are pragmatically motivated and those that are not, respectively.
With respect to the issue of the relation between discourse and syntax,
RRG differs fundamentally from the syntactocentric approaches to linguis-
tic theory. On the one hand, it contrasts with Chomsky’s (1995) Minimalist
Program, which, at least in its original formulation, denied any status to
discourse in the syntactic computation, strictly associating discourse with
interface effects. On the other hand, it diverges from Cartography (Rizzi
1997, 2006), where movement operations targeting pragmatic positions start
c. French
«(Ma voiture) elle est en panne.»
my car it be3sg in breakdown
d. Japanese
«(Kuruma-wa) koshoo-shi-ta.»
car-top breakdown-do-pst
(16) Mandarin (Van Valin and LaPolla 1997: 207, from Li and Thompson 1976)
Nèi xiē shù, shùshēn dà.
that few tree trunk big
‘Those trees, the trunks are big.’
(17) Students, you have to remind them of the deadlines for coursework
submission every second week.
The topic of the utterance in (17) is a detached information unit in the pre-
detached position (see Figure 11.4), whereas the rest of the sentence is a
syntactic core (you have to remind them of the deadlines for coursework submission)
with an additional peripheral adverbial (every second week), both being within
the domain of the assertion.
As should be clear from the above, a defining feature of predicate focus
is the availability of the subject matter at issue from the presupposition.
Another defining feature of predicate focus can be explained with respect
to an important distinction made by RRG, that is the contrast between
the potential focus domain, which is the syntactic domain in the sen-
tence in which focus can occur in a given language, and the actual focus
domain, which is the syntactic constituent or constituents of a given
sentence that is or are in focus. In light of this distinction, predicate
focus is characterized by the limitation of the actual focus domain to the
syntactic core minus the subject-topic. This is shown in the speech act
projection in (15a), where the continuous line indicates the actual focus
domain, while the dotted line indicates the potential focus domain in
English.
Sentence focus differs from predicate focus insofar as the matter at issue
is not introduced previously in discourse, the whole proposition is asserted,
and the whole sentence is in the actual focus domain. Given that they do not
have a presupposed topic, sentence-focus structures occur out of the blue.
Lambrecht (1994: 223), the source of the following examples, also claims
that they reply to the question what happened?
The fact that the argument is in focus is reflected in the prosody of the
English construction in (18a) and in the morphosyntax of the constructions
in (18b–d). In both (18b) and (18c), the focal argument occurs postverbally.
Since French poses strict constraints on verb–subject order, the focal
argument is not encoded as a subject but rather as the undergoer of avoir
‘have’. Verb–subject order can also be found in English, although a locative
phrase or dummy there is required preverbally (There arrived three buses at the
station) and there are verb-class restrictions on this construction (Levin and
Rappaport Hovav 1995: 215–277). In Japanese, the focal argument is
marked with the nominative marker ga. The same focus meanings can
thus be expressed by different prosodic, morphological and syntactic struc-
tures in different languages.
Given that, in sentence focus, information is not structured as a contrast
between presupposed and non-presupposed information, this type of focus
structure expresses thetic statements, that is, unstructured statements
which describe a situation or an event, rather than making a statement
about a previously introduced individual or entity. Following Lambrecht
Although bought a car is the comment on the topic he, only a car is in the
actual focus domain, since it is in narrow focus.
However, since the potential focus domain extends to the whole clause, in
English, and focus is primarily marked prosodically, it can also fall in
marked positions, as is shown in (24).
It should be noted, however, that narrow focus can be contrastive even when
it occurs in an unmarked position.
Contrastiveness as a relational information structure property that
selects from alternatives is orthogonal to the distinction between presuppos-
ition and assertion, in that the alternatives can be stated or predicted as part
of the presupposition. Thus, not only can foci be contrastive, but topics can,
too.
and foci can both be marked with -wa, although by default contrastive foci
are marked with -ga, and contrastive topics with -wa (see Shimojo 2009,
2010, 2011 for extensive discussion). To capture the topic–focus ambivalence
in topic marking in Japanese, Shimojo (2011) subsumes Erteschik-Shir’s
(1997, 2007) notion of subordinate f(ocus) structure within the RRG frame-
work. A subordinate f-structure is a complex f-structure involving embed-
ding. Contrastive information units can be topics or foci because there is a
contextually available set, which is topical, whereas the contrastive infor-
mation units that are selected from this set are foci. A structure with
contrastive -wa is thus represented as follows by Shimojo (2011: 275).
In terms of sentence structure, Shimojo (2010, 2011) proposes that the -wa-
marked constituent figures outside the domain of focus in syntax, specific-
ally in the pre-detached position.12
Since typological variation is a key concern of RRG, attention has been paid
to the cross-linguistic variation in the interplay of prosody and syntax in the
expression of information structure (Van Valin 1999). In some languages, the
three types of focus structure discussed in Section 11.4 are primarily
encoded in syntax by means of word order, whereas in others they are
mainly encoded prosodically. In this second type of language, focus struc-
ture tends to be syntactically flexible, insofar as the potential focus domain
is not restricted to a particular portion of the clause.
Van Valin (1999) discusses this kind of typological variation in terms of
the relative rigidity or flexibility of focus structure and syntax across
languages. French, Toba Batak (a Western Austronesian language spoken
in Indonesia) and other languages have rigid syntax and rigid focus struc-
ture. This means that while the potential focus domain (see Section 11.4)
does not extend to the whole clause, focus cannot be freely marked by the
syntactic position of the focal unit either, since the syntactic position of
arguments is fixed to various degrees. In French, the actor of an active
clause occurs in the core-internal immediately pre-nuclear position. In the
spoken registers, it is usually omitted, or detached outside the clause,
while a coreferent clitic pronoun is spelled out preverbally. We shall not
dwell here on the status of this pronoun in morphosyntax, namely
whether it must be assumed to occur in a pre-nuclear position or, rather,
under the nuclear node (see Belloro 2004, 2015 for the position of core
clitic arguments in RRG). The point that is relevant in the current discus-
sion is that the actor of the active voice cannot normally occur in post-
nuclear position in French. Using terminology adopted in other frame-
works, French disallows free subject inversion (though see Lahousse 2011
and Lahousse and Lamiroy 2012 for further discussion). Focus, in turn, is
avoided in the core-internal immediately pre-nuclear position and is
marked post-nuclearly by default.13 The potential focus domain in the
French clause is thus as shown in (28).
(28) French
[PrDP Marie] [ Clause [ Core elle [ Nuc aime] les bonbons]] ]
Mary she love.3 SG the sweets
Wh-words, which are focal, can occur preverbally, though in the pre-core
slot, rather than in a core-internal position. However, they can also occur in
situ, thus taking a post-nuclear position (cf. 29a–b). Otherwise, they are
clefted (cf. 30a).
(29) French
a. Vous avez vu qui?
you.2pl have.2pl seen who
‘Who have you seen?’
b. Vous allez o ù?
you.2pl go.2pl where
‘Where are you going?’
Given that the extent of the potential focus domain rules out focal
subjects in situ, while the syntax of French normally bans post-nuclear
subjects, the violation of these pragmatic and syntactic restrictions is
avoided by clefting (cf. 30b) or by using subjectless VS constructions (cf. 31).
(30) French
a. C’est qui qui a fait ça?
it be.3sg who who have.3sg done this
‘Who has done this?’
b. C’est Marie qui a fait ça.
it be.3sg Mary who have.3sg done this
‘It’s Mary who did this.’
(31) French
Il s’ est produit des problèmes.
it refl be.3sg produced some problems
‘There occurred some problems.’
In Toba Batak, which is a language with rigid VOS order and immediately
post-nuclear focus, the verb must carry a voice prefix when the undergoer is
topical and the actor is not (Van Valin 1999). This suffix indicates that
the undergoer has privileged syntactic argument status: the syntactic
(33) Italian
[ Core [ Nuc Sono emersi] tanti problemi ].
be.3 PL emerged many problems
(34) Italian
[PrCS Chi] [Core [Nuc ha fatto] questo]?
who have.3sg done this
‘Who did this?’
To return to Italian, not only does the pre-core slot host wh-words, but it also
admits contrastive narrow foci. This type of focus fronting is shown in (38b).
(38) Italian
a. «Cosa ha comprato Giorgio? La macchina?»
what have.3sg bought George the car
‘What did George buy? A car?’
b. «La bicicletta, Giorgio ha comprato, non la macchina.»
the bike George have.3sg bought neg the car
‘A bike George bought, not a car.’14
The fact that, in (38b), the focal argument co-occurs with the actor in pre-
nuclear position indicates that the position of the contrastive focus is core-
external, specifically, the pre-core slot (see Bentley 2008 for a fuller discussion
of this point). Observe that the preverbal actor must be assumed to occur core-
internally, specifically in the core-initial position. Indeed, RRG rules out the
assumption that a topic can follow the focus in the pre-core positions, that is,
those positions which, in other frameworks, are referred to as the left periph-
ery of the clause (Van Valin 2008: xx in disagreement with Rizzi 1997).
Languages with flexible syntax and flexible focus structure extend the
potential focus domain to the whole clause. Van Valin (1999) provides
relevant examples from Slavic, which are reported here.
The fact that both SV and VS order are allowed in sentence focus sets Russian
apart from the languages discussed previously, suggesting that preverbal
focus is not constrained in Russian in the same way as it is in Italian.
It should be noted, however, that the topic–focus order is adhered to in
declarative clauses with focus on the postverbal argument, while wh-words
occur in the pre-core slot (Van Valin 1999, based on Comrie 1979, 1984).
(42) Italian
a. «Che ha comprato Giuseppe a suo nipote?»
what has buy.ptcp Joseph to his nephew
‘What did Joseph buy for this nephew?’
b. «*La bicicletta ha comprato.»
the bike has buy.pst.3sg
‘He bought him a bike.’
Data like (42)–(43) differentiate Sicilian from Italian, indicating that there
is no ban on core-internal pre-nuclear focus in Sicilian, and that the latter
language is more flexible than the former in terms of its focus structure.
Sicilian pre-nuclear focus, however, is not entirely free, but rather has
affective and emphatic connotations, which are absent from post-nuclear
focus (Sornicola 1983; Leone 1995; Cruschina 2006). In particular, preverbal
Narrow focus can fall in any position in the English clause (cf. 46b, 47a–d)
and is marked prosodically.
Once the logical structure of the predicate is retrieved from the lexicon,
its argument positions are filled with the core arguments, which are coded
in accordance with their pragmatic states and their roles as part of the
presupposition or the assertion (see Figures 11.2 and 11.3). Thus, in English,
the activated topical subject of a predicate-focus structure would most likely
be an unstressed pronoun, while in Italian and in the languages that allow
phonologically null subjects it would most likely not occur at all.
(50) Italian
Parl-o.
speak-1sg
‘I speak.’
do′ (1sg act , [speak′ (1sg act )])
However, when the actor is focal and the undergoer is not, the former is
marked by contrastive ergative morphology on pronominal actors and by
ergative case on non-pronominal ones, regardless of the animacy of the
undergoer.
Finally, in languages with flexible syntax, narrow focus and sentence focus
may involve marked word order patterns, for example VS, in languages with
default SV order (Section 11.5). Again, this is reflected in the choice of a
syntactic template for the clause, as is the case with the Russian example of
sentence focus previously given in (39c) and repeated below for convenience.
(57) Russian
[Core [Nuc Slomalac’] mašina].
broke.down car
‘[My] car broke down’
Prosody also plays a role in the linking, spelling out focus and providing
clues on the domains of presupposition and assertion in the processing of
the sentence. As was mentioned in Section 11.5, it has been proposed that
the role of prosody in information structure ought to be represented in a
prosodic projection (O’Connor 2008).
Although in this section we have discussed the role of discourse in the
linking from semantics to syntax, which pertains to language production,
we should note that discourse also plays a key role in the linking from
syntax to semantics, which pertains to language comprehension. By way of
conclusion of the section, we will briefly touch upon this direction in the
linking with reference to zero anaphora.
Zero anaphora is puzzling because it would seem to challenge the Com-
pleteness Constraint (Van Valin and LaPolla 1997: 325), which states that all
the arguments explicitly specified in the semantic representation of a sen-
tence must be realized syntactically in that sentence. In Mandarin Chinese
and other languages, a presupposed argument can be entirely silent (Van
Valin 2005: 174). Differently from the silent privileged syntactic argument of
many pro-drop languages (see the Italian example in (50)), the silent argu-
ment under discussion here has no pronominal or affixal expression on the
verb. Another puzzling case of zero anaphora is that of the Japanese verb-less
numeral quantifier construction, where an argument bears accusative case,
but there is no overt case assigner in clause structure. In this case, syntax
lacks a verb rather than an argument.
RRG captures these patterns of zero anaphora by linking the semantic
representation directly with discourse, drawing upon Kamp and Reyle’s
(1993) Discourse Representation Theory. Proper discussion of the application
of this theory to RRG is provided in Chapter 12, where Van Valin and
Latrouite propose an Extended Completeness Constraint, which states that
completeness can be satisfied by discourse representation structures in
combination with arguments that are expressed overtly in syntax. Here we
note that discourse representation keeps track of presupposed and asserted
information, which can be retrieved directly in the semantics of the clause,
without being expressed in syntax. Thus, the silent arguments of Mandarin
Chinese can be filled into the semantic representation in syntax–semantics
linking because they can be drawn from discourse representation, as can the
11.7 Conclusion
Some of the most original and significant results of research in RRG suggest
that an important way in which languages differ from each other is in terms
of the role of discourse in the linking of semantics with syntax and syntax
with semantics. RRG offers an approach to information structure which is
flexible enough to capture this variation, while also being sufficiently con-
strained to make important generalizations regarding the expression of
pragmatic states and pragmatic relations, and their interface with prosody,
morphology and sentence structure across languages. In this chapter we
have introduced the constructs adopted in the RRG approach to information
structure and we have discussed the role that this framework gives to
information structure in the architecture of grammar.
References
Belloro, Valeria. 2004. A Role and Reference Grammar Account of Third Person Clitic
Clusters in Spanish. MA thesis, University at Buffalo (SUNY).
Belloro, Valeria. 2015. To the Right of the Verb: An Investigation of Clitic Doubling
and Right Dislocation in Three Spanish Dialects. Newcastle upon Tyne: Cam-
bridge Scholars Publishing.
Bentley, Delia. 2008. The interplay of focus structure and syntax. Evidence
from two sister languages. In Van Valin, Jr. (ed.), 263–284.
Bentley, Delia. 2010. Focus fronting in the layered structure of the clause. In
Nakamura (ed.), 4–28. https://rrg.caset.buffalo.edu/rrg/ProceedingsofRRG2009_
02.pdf.
Bentley, Delia. 2018. Grammaticalization of subject agreement on evidence
from Italo-Romance. Linguistics 56: 1245–1301.
Bentley, Delia, Francesco Maria Ciconte, Silvio Cruschina and Michael Ram-
sammy. 2016. Micro-variation in information structure: Existential con-
structions in Italo-Romance. In M. M. Jocelyne Fernandez-Vest and Robert
D. Van Valin, Jr. (eds.), Information Structure and Spoken Language in a Cross-
Linguistic Perspective, 95–120. Amsterdam: John Benjamins.
Gundel, Jeanette K., Nancy Hedberg and Ron Zacharski. 1993. Cognitive
status and the form of referring expressions in discourse. Language 69(2):
274–307.
Jackendoff, Ray. 2002. The Architecture of the Language Faculty. Cambridge, MA:
MIT Press.
Kamp, Hans and Uwe Reyle. 1993. From Discourse to Logic. Hingham, MA:
Kluwer.
Krifka, Manfred. 2007. Basic notions of information structure. In Caroline
Féry, Gisbert Fanselow and Manfred Krifka (eds.), The Notions of Information
Structure, 13–55. Potsdam: Universitätsverlag.
Kuno, Susumo. 1972. Functional sentence perspective: A case study from
Japanese and English. Linguistic Inquiry 3: 269–320.
Lahousse, Karen. 2011. Quand passent les cicognes: Le sujet nominal postverbal en
français moderne. Paris: Presses Universitaires de Vincennes.
Lahousse, Karen and Béatrice Lamiroy. 2012. Word order in French, Spanish
and Italian: A grammaticalization account. Folia Linguistica 46: 1–29.
Lambrecht, Knud. 1986. Topic, Focus and the Grammar of Spoken French. PhD
dissertation, University of California, Berkeley.
Lambrecht, Knud. 1994. Information Structure and Sentence Form: Topic, Focus,
and the Mental Representation of Discourse Referents. Cambridge: Cambridge
University Press.
Lambrecht, Knud. 2000. When subjects behave like objects: An analysis of
the merging of S and O in sentence focus constructions across languages.
Studies in Language 24: 611–682.
LaPolla, Randy J. 1990. Grammatical Relations in Chinese: Synchronic and Dia-
chronic Considerations. PhD dissertation, University of California, Berkeley.
LaPolla, Randy J. 1993. Arguments against ‘subject’ and ‘direct object’ as
viable concepts in Chinese. Bulletin of the Institute of History and Philology 63
(4): 759–813.
LaPolla, Randy J. 1995. Pragmatic relations and word order in Chinese. In
Pamela Downing and Michael Noonan (eds.), Word Order in Discourse,
297–329. Amsterdam: John Benjamins.
Leone, Alfonso. 1995. Profilo di sintassi siciliana. Palermo: Centro di Studi
Filologici e Linguistici Siciliani.
Levin, Beth and Malka Rappaport Hovav. 1995. Unaccusativity: At the Syntax–
Lexical Semantics Interface. Cambridge, MA: MIT Press.
Li, Charles N. and Sandra Thompson. 1976. Subject and topic: A new typ-
ology of language. In Charles Li (ed.), Subject and Topic, 459–489. New York:
Academic Press.
Nakamura, Wataru (ed.). 2010. Proceedings of the 10th International Conference on
Role and Reference Grammar (RRG 2009). https://rrg.caset.buffalo.edu/rrg/Pro
ceedingsofRRG2009_02.pdf.
O’Connor, Rob. 2008. A prosodic projection for Role and Reference Gram-
mar. In Van Valin, Jr. (ed.), 227–244.
Van Valin, Robert D., Jr. 1999. A typology of the interaction of focus structure
and syntax. In Ekatarina Raxilina and Yakov G. Testelec (eds.), Typology and
Linguistic Theory: From Description to Explanation, 511–524. Moscow: Lan-
guages of Russian Culture.
Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cam-
bridge: Cambridge University Press.
Van Valin, Robert D., Jr. (ed.). 2008. Investigations of the Syntax–Semantics–
Pragmatics Interface. Amsterdam: John Benjamins.
Van Valin, Robert D., Jr. 2015. An overview of information structure in three
Amazonian Languages. In M. M. Jocelyne Fernandez-Vest and Robert
D. Van Valin, Jr. (eds.), Information Structuring of Spoken Language from a
Cross-linguistic Perspective, 77–92. Berlin: De Gruyter Mouton.
Van Valin, Robert D., Jr. and Randy LaPolla. 1997. Syntax: Structure, Meaning
and Function. Cambridge: Cambridge University Press.
Notes
frame setters and contrastive topics should take the same syntactic
position, although some evidence in support of this assumption is
reported in Frascarelli (2017: 486).
8 This notion of focus draws upon Lambrecht’s (1994: 213) claim that focus
is ‘the semantic component of a pragmatically structured proposition
whereby the assertion differs from the presupposition’.
9 The former names Left- and Right-Detached Position reflected a bias in
favour of left-to-right writing systems, differently from the new termin-
ology, which is based on order of occurrence in speech.
10 In the Gallo-Italian dialects of Northern Italy, the topic of sentence-focus
constructions with verbs of specific lexical classes can be expressed
overtly by a locative clitic in subject clitic position (Tortora 1997, and,
for an RRG account, Bentley 2018). The question arises of whether this
locative clitic should take a syntactic position. This question, however,
goes beyond the scope of the present chapter.
11 It should also be noted that the VP does not have any status in the
layered structure of the clause, because not all languages provide evi-
dence for it. Therefore, focus on the core minus the subject-topic cannot
be defined as focus on the VP constituent in RRG.
12 Further evidence that contrastive topics and foci can be marked in the
same way in individual languages has been found in Romance dialects
spoken in the north-east of Italy (De Cia 2019).
13 See Stempel (1981) for some deviations from this strong tendency
of French.
14 The definite marker in the RP can be said to be generic and it is not a
marker of specificity or uniqueness. Thus, this example could be trans-
lated as ‘now George is a bike owner, not a car owner’. One could also
find an indefinite RP in this example (Una bicicletta Giorgio ha comprato,
non una macchina ‘A car George has bought, not a car’), with the same
focus structure as discussed above.
15 Core-initial focus occurs in out-of-the blue ascriptive structures, where
the focus is on a non-verbal predicate (Sic. Stanca sugnu, lit. tired I-am).
16 The abbreviations contr and tns in the glosses of examples (51)–(53)
stand for ‘contrastive’ and ‘tense’, respectively.
Abbreviations
12.1 Introduction
Bentley introduced the distinction between the potential and actual focus
domains and the notion of information units, which are very important
parts of the RRG account of IS, along with representation of them in the LSC.
The LSC discussed so far has two projections, the constituent projection and
the operator projection. The theory posits two more, one for IS and one for
prosody (O’Connor 2008). They are treated as distinct projections, because
not all languages use prosody to signal IS (Van Valin 2016), even though
many, if not most, do.
The essential components of the focus structure projection as presented in
Van Valin (2005) are: (i) basic information units (IU), (ii) the potential
focus domain (PFD), and (iii) the actual focus domain (AFD). These are
illustrated in Figure 12.1.
The opposition between the PFD and the AFD is central to the RRG
analysis of IS. The PFD is a function of the grammar. In English the PFD
in simple clauses encompasses the entire clause, as can be readily seen in
SENTENCE
CLAUSE
CORE
RP NUC RP PP
PRED
Actual focus domain
V
Chris presented a child with some flowers
SPEECH ACT
Figure 12.1 Components of the focus structure projection (Van Valin 2005: 77)
Figures 12.1 and 12.2. In Italian, on the other hand, it excludes the pre-
nuclear core-initial position, as Bentley discusses in some detail (see also
Bentley 2008). The AFD is a function of the context in which the sentence
occurs; it is the focus of the utterance. It corresponds to the ‘focus domain’
of Lambrecht.
There is a very important IS distinction which is not represented here,
namely topic vs. comment. According to both Gundel (1988) and Lam-
brecht (1994), the topic referent is part of the pragmatic presupposition,
which is part of the immediate common ground (ICG) (Berio et al. 2017;
Krifka and Musan 2012). The ICG consists of the discourse context, which
includes the pragmatic presuppositions, plus the immediate social and
physical environment in which the speech event occurs. World knowledge
is contained in the general common ground (GCG). In Figure 12.1, there
is only a single IU outside of the AFD, namely, the RP Chris, which is
interpreted as the topic expression which refers to the topic referent. This
is not always the case, however, as it is possible for there to be more than
one IU outside the AFD, as illustrated in Figure 12.2.
The utterance represented in Figure 12.2 could be the answer to the
question What did John buy?, which establishes the referent of John and the
action of buying as part of the pragmatic presupposition in the ICG, and
the AFD is a new car. The part of the PFD which does not include the AFD is
termed the background.1 The individual referred to by the RP John is
presumably the topic, but this is not indicated in the representation. Balogh
(2021) proposes to augment the focus structure representation with nota-
tions indicating topic and comment. Following her proposal, the revised
representation would be as in Figure 12.3. Since it now codes both focus–
background and topic–comment, she labels it the ‘information structure
projection’.2
Figure 12.4 The constituent, operator and information structure projections of a sentence
v w, x y, z
respectively. Since ‘w’ is a pronoun, its reference must be specified, and the
fact that it refers to ‘v’ is stated, and then the proposition ‘w opened x’ is
given. The final sentence is uttered in the context created by the first two
sentences, so the information from them is carried over, and the informa-
tion from the first DRS is light grey and that from the second is dark grey,
indicating their relative salience in the ICG. It introduces two more refer-
ring expressions, Mary (assigned ‘y’) and him (assigned ‘z’). Him being a
pronoun, its reference must be specified as ‘z ¼ w’, and the proposition in
it is ‘y kissed z’.
How are focus structures related to the DRSs? In Lambrecht’s approach the
focus is defined as the assertion minus the presupposition. Consider the
following question–answer pairs, in which the focus is in small caps.
The (b) sentence is structurally the same in both pairs, but the focus is
different, due to the different presuppositions in the ICG established by
the questions. This can be represented as shown in Figures 12.6 and 12.7.
feature of this construction, and without the numeral classifier the clause
with the copula and the accusative marking on the undergoer is ungram-
matical, unless it is an emphatic utterance to focalize ringo ‘apple’ (M.
Shimojo, pers. comm.).
The central question this construction poses is, ‘what licenses the accusa-
tive case on the RP in the absence of a transitive verb?’ A second question
concerns the interpretation of the actor in the sentence. It is a general
principle of Japanese that continuing, highly activated elements of the
ICG, especially topical actors (PSAs), are not normally expressed overtly.
Hence representing the activated constituents is crucial for explaining this
construction. Shimojo did an internet search for instances of this construc-
tion and looked at the first hundred examples. In eighty-nine of them the
actor and the verb were recoverable from the previous utterance; in eight of
them the interpretation was predictable from the preceding context (e.g. ‘I
just went to Starbucks. Two tall lattes-acc is,’ where the most likely mean-
ing is ‘bought’ or ‘drank’). The final three cases were online classified ads,
where the only verb that fits the context is ‘sell’. Thus, the interpretation of
the verb-less numeral classifier construction depends on either the immedi-
ately preceding utterance or contextually relevant sociocultural background
knowledge (the GCG). Shimojo’s RRG analysis based on representing context
via DRSs works for the eighty-nine discourse-context-dependent examples.
The preliminary step in the analysis in terms of linking from semantics to
syntax is to represent the activated information in the ICG; in this example,
it is ‘Taro got a snack’.5
The first two steps in the linking from semantics to syntax are to get the
logical structure (LS) of the predicate from the lexicon along with the
referring expressions that fill the argument positions, based on the message
to be expressed, first, and assign the semantic macroroles, second. This
yields Figure 12.8b.
The next step is to select the PSA and assign case. This yields Figure 12.8c.
This looks like a standard linking with a transitive verb in any accusative
language. But now the general principle that highly activated continuing
elements in the ICG are not normally expressed overtly in Japanese comes
into play. The next step involves the selection of the syntactic templates from
the syntactic inventory. The basic principle states that the number of argu-
ment slots in the core template corresponds to the number of specified
argument positions in the semantic representation, as a default, and it also
assumes a slot for the predicate in the nucleus. Based on these representa-
tions, it appears that the syntactic template should have two slots for core
arguments, but because the actor-PSA and the verb are part of the presuppos-
ition, they don’t figure in the determination of the core template. The only
part which is not part of the presupposition is ‘two apples’, and so this results
in the selection of a core template with only one argument slot (Figure 12.8d).
The final step is linking the accusative undergoer ‘two apples’ to the
argument position and inserting the copula in the nucleus to carry the
Figure 12.8e Summary of linking from semantics to syntax in Japanese verb-less numeral quantifier
construction in (3b)
Figure 12.9 Summary of linking from syntax and discourse to semantics in Japanese verb-less
numeral quantifier construction
logical structure of the verb identified in the previous step. But there is no
predicating verb in the sentence, and accordingly it is necessary to refer
again to the background DRS for it. The logical structure for ‘GET’ is called
up and its two arguments are assigned actor and undergoer macroroles.
The final step is the association of the arguments in the DRS and the syntax
with the arguments in the logical structure: Taro is the x-argument and the
actor of ‘GET’, and the a argument is the actor, hence Taro links to the a
argument; similarly, ringo-o ni-ko is an undergoer, and the b argument is
the undergoer in the logical structure of ‘GET’, therefore ringo-o ni-ko links
to the b argument. The resulting interpretation is given below the asser-
tion DRS. Here again the Extended Completeness Constraint is satisfied by
arguments from a DRS and the syntax.
This construction is particularly interesting, because it requires context
not only for the recovery of an argument, which is the usual situation with
discourse-driven zero anaphora, but also for the interpretation of the predi-
cate (verb). This is a phenomenon which is not limited to languages like
Japanese. English provides an example with ‘VP’-ellipsis across speakers, as
exemplified in (4b, b′).
(4) a. Kim is eating an ice cream cone, and Pat is, too.
a′. Kim is eating an ice cream cone, and so is Pat.
b. Speaker 1: Kim is eating an ice cream cone.
Speaker 2: Pat is, too.
b′. Speaker 2: So is Pat.
In Van Valin (2005: 231–235) the linking in (4a, a′) was presented in order to
show how RRG handles ‘VP’ phenomena despite not positing a VP node in
the constituent projection of the clause. Key to the analysis is the projection
of the semantic representation of the first clause onto the second clause in
the sentence, and this is possible because the two clauses are part of a single
sentence produced by a single speaker. This is not the case in (4b, b′),
however, as the two clauses are produced by two separate speakers in
separate utterances. Yet the interpretation is the same regardless of whether
the clauses are part of the same utterance or two utterances.
How can the interpretation be accounted for in the multi-speaker situ-
ation? The answer, as with the Japanese verb-less nominal classifier construc-
tion, involves DRSs. Instead of copying the semantic representation of the
first clause resulting from the syntax-to-semantics linking (with the PSA
replaced by a variable) onto the second clause, the result of the linking
constitutes the assertion made by the utterance, which is represented by a
DRS, and it becomes part of the ICG and functions as the pragmatic presup-
position for the second assertion. Because it is necessary for the construc-
tion, the DRS contains the tense–aspect values of the clause. The proposition
and operators contained in this DRS are projected as the semantic represen-
tation of the second utterance, and the single RP in the second utterance is
(5) a. I expected to find someone mowing the yard, and mowing the yard was Bill.
a′. #I expected to find someone mowing the yard, and mowing the yard Bill was.
b. I expected to find Bill mowing the yard, and mowing the yard he was.
b′. #I expected to find Bill mowing the yard, and mowing the yard was he.
In the (5a) examples, the first clause does not introduce a specific referent for
the actor of mow, and when the referent is introduced, it is more felicitous to
have it after the finite auxiliary verb rather than before it; what we have here
is in effect ‘focal subject inversion’, rather like what is found in Italian and
discussed by Bentley in Chapter 11. When, on the other hand, the first clause
introduces a specific referent, as in the (5b) examples, the pronoun referring
12.5.1 Step 1
According to the model of a speaker proposed in Levelt (1989), the first step in
the production of an utterance is the formulation of an idea or thought to be
communicated, and the initial fully linguistic step is to convert it into a
semantic representation, which is step 1 in (6). This is primarily semantically
driven, but IS can have an influence in a number of ways. First, certain focus
constructions favour verbs of certain types. A good example is sentence focus
constructions, which have robust contextual constraints on their occurrence
(cf. Bentley, Chapter 11). Lambrecht (1987), along with Kuno (1972a), observed
that ‘the predicates most commonly permitted in [sentence focus] sentences
involve “presenting verbs”, i.e. intransitive verbs expressing appearance or
disappearance of some referent in the internal or external discourse setting,
or the beginning or end of some state involving the referent’(1987: 373).
A second example comes from languages without productive voice
systems. It was discussed in Chapter 5 on grammatical relations that IS
can influence PSA selection in languages with voice operations (see also
Section 12.5.3). It doesn’t follow, however, that IS is completely lacking in
influence in languages without voice. Languages may possess verbs which
present a state of affairs from different perspectives, as illustrated in (7).
These pairs of sentences have different verbs which reflect different PSA
choices from among the participants in the state of affairs, and the selection
of a verb with its PSA reflects the salience of the referent of the PSA in the
discourse. In a context in which the referent of Bill is the topic and the
referent of Sally is newly mentioned, (7a, c′) would be more natural than (7a′,
c). Similarly, if the topic is the pandemic, then (7b) is appropriate, whereas if
Bill’s cousin is being talked about, then (7b′) is perhaps more natural. In a
language without voice, the primary way to vary the PSA in terms of the
different topicality statuses of the participants is through selecting a verb
which describes the same state of affairs as another verb but with a different
PSA, as in (7).6
A third, and very important, case of IS influence concerns the coding of
referring expressions. In (7a), for example, the referent of the giver of the
book is coded as a proper name, Bill. There are, however, a wide variety of
options available to a speaker, depending on the status of the referent in the
ICG: someone, a man, this guy, this/that man, the man, the man I met at the party
yesterday, the jerk/fool/idiot/genius, Bill, Mr Smith, he, that one, . . . All of these
different referring expressions can be used to denote the individual ‘Bill
Smith’, and the choice is a function of a number of factors, such as how
recently he has been referred to, how familiar the speaker believes the
interlocutor is with him. (See Bentley’s chapter on IS, Section 11.2, for a
detailed introduction to these distinctions.) In some languages, especially
head-marking languages, coding options include bound morphological pro-
nouns and anaphors, and clitics. The inventory of coding possibilities varies
from language to language, but in every language IS factors influence the
selection in actual utterances.
12.5.2 Step 2
The second step in (6) is macrorole assignment. In terms of the AUH, the actor
is the leftmost specified argument in the LS, and the default choice of under-
goer is the rightmost specified argument in the LS. This was shown in the
discussion of macrorole choice in Chapters 1 and 4. With respect to actor
selection, the only variation is whether the leftmost argument will be speci-
fied, that is, morphosyntactically instantiated in the clause, or left unspeci-
fied, as in an ‘agent-less’ passive, for example. In some cases this may be
motivated by the activation status of the referent of the leftmost argument, as
we saw in the Japanese example in (3b) in which highly activated PSA-
referents need not be explicitly mentioned again. In sentences involving
causal chains with two effectors, for example Sam opened the chest with the
skeleton key, the instigating animate effector can be left unspecified and the
intermediate inanimate effector, the instrument, can be selected as the actor,
yielding The skeleton key opened the chest.7 In both of these versions the actor is
the leftmost specified argument in the LS of the verb.
(8) a. Dative shift: Sam handed the letter to Alice vs. Sam handed Alice the letter.
b. Transfer alternation: Max presented the prize to Sally vs. Max presented Sally
with the prize.
c. Locative alternation: Mary sprinkled sugar on the cookies vs. Mary sprinkled the
cookies with sugar.
d. Removal alternation: Larry drained the water from the pool vs. Larry drained the
pool of its water.
Each of the questions establishes the referent of either the recipient or the
theme as the topic in this limited context, and the more natural response to
it is the one in which the topic expression precedes the newly mentioned
referent. Thus, it is reasonable to conclude that there are cases in English
where IS can influence undergoer selection. It remains to be seen if there is
any evidence that this happens in other languages. German and Croatian,
for example, have limited variable undergoer selection, and it is doubtful
that it could have the same IS motivation as in English, since, as noted
earlier, simply reversing the order of dative and accusative RPs achieves the
same effect.
12.5.3 Step 3
The third step in the semantics-to-syntax linking is the assignment of mor-
phosyntactic functions and properties. The former concerns primarily PSA
selection and the latter agreement and case assignment. The influence of IS
on PSA selection in some languages was discussed in the chapters on
grammatical relations (5) and IS (11), and we will present a particularly
striking example of it. Ever since Schachter (1976), Tagalog has been seen
as a language in which IS plays a role in PSA selection. While this is often
discussed in terms of the definiteness or specificity of the undergoer being
crucial for PSA and voice selection (see LaPolla Chapter 5), there are add-
itional factors at work, as Latrouite (2011) shows. For example, causative
change-of-state predicates (e.g. tatay ‘kill’) strongly favour undergoer voice,
whereas activity predicates favour actor voice. Latrouite call this ‘event
structural prominence’, and it interacts with IS prominence in the deter-
mination of voice and PSA selection. A third factor is whether the actor or
undergoer argument is displaced to the beginning of the utterance.
Latrouite and Van Valin (2021), building on Schachter and Otanes (1972),
survey displacement constructions in Tagalog, and the two that are most
relevant to this discussion are ang-inversion and ay-inversion; the former is
illustrated in (12) with the verb sulat ‘write’.8
(12a, b) represent the basic word order in a simple clause in Tagalog; (a)
illustrates actor voice, (b) undergoer voice. The nucleus is the first element
in the clause, and the default position for the ang-marked PSA is core-final.
The ang-inversion construction in (c) and (d) is usually translated by an
English it-cleft, which indicates that it should be interpreted as narrow focus
(see Latrouite 2021). In these examples the inverted RP is the nominative RP
congruent with the voice of the predicator in the nucleus, that is, in (c) the
verb is actor voice, and the displaced RP is the actor, while in (d) the verb is
undergoer voice, and the inverted RP is the undergoer. However, this correl-
ation does not always hold: Latrouite and Riester (2018) found that there
were cases where the actor was inverted but the verb was undergoer voice, as
exemplified in (13a). It is, however, not possible to invert the undergoer in
actor voice, as the ungrammaticality of (13b) shows.
What explains this asymmetry? Latrouite and Riester argue that actor and
undergoer arguments have default IS values, with the actor being the
default topic and the undergoer being the default focus. With a topical
actor and an undergoer as part of the focus, IS does not require special
marking, and referentiality and verb semantics influence voice selection. If
the IS mapping is divergent, for example if the actor is topical (unmarked)
and the undergoer is topical (marked), we get undergoer voice to signal this
divergence. If the undergoer is a contrastive displaced topic (marked) and
the actor is topic (unmarked), we will get undergoer voice. If the actor is
focal (marked) and the undergoer is focal (not marked), then actor voice and
displacement (in the case of contrastive focus) tends to be used for
this divergence.
These default correlations are not surprising, since actors are the default
choice for PSA in the vast majority of languages (see Van Valin and LaPolla
1997, §6.5), and the default focus structure, predicate focus, has the PSA as
topic and the undergoer as focal. What is surprising is that IS can override
morphosyntactic constraints: focal actors are strongly preferred in displaced
(inverted) positions, as in (13a); note that the undergoer is the PSA and
therefore topical, which is the non-default IS function of undergoers. Hence
a displaced focal actor with a topical undergoer PSA and undergoer voice
reflects the marked IS-functions of the arguments. Why is (13b) impossible?
Two factors may be at work. First, the ang-marked lowest-ranking argument
of the verb is normally non-referential, which makes it a poor candidate for
displacement. Moreover, ang-marked arguments are always referential.
Hence the occurrence of ang-marking on the lowest-ranking argument is
incompatible with actor voice. In (12d), on the other hand, undergoer voice
signals that the lowest-ranking argument is referential, and therefore ang-
marking is appropriate. Thus one of the functions of the construction in
(12d) is to express a focal referential undergoer. This is a complex and
interesting example of the interplay of semantic macroroles and IS func-
tions with PSA and voice selection.
Agreement between the verbal complex in the nucleus and one or more
core arguments and case assignment shows an interesting asymmetry:
there appear to be many more examples of IS-influenced case assignment
than of IS-influenced agreement. The main examples of IS-influenced
finite verb or auxiliary agreement concern sentence focus constructions
in languages with restrictions on the PFD in simple clauses. It was pointed
out in Chapter 11 that in Italian the PFD in simple clauses excludes the
pre-nuclear position in the core; the AFD is restricted either to the pre-
core slot (PrCS) or to the nucleus plus post-nuclear constituents. Hence no
core-internal preverbal focus is permitted, and in the case of sentence
focus, the focal ‘subject’ must occur after the nucleus. It was also men-
tioned that in the southern Bantu languages Setswana and Sesotho
(Demuth 1989), the restriction is even stricter: no pre-nuclear focal mater-
ial is permitted at all. In sentence focus the focal argument must be post-
nuclear, and the predicate in the nucleus fails to show agreement. The
contrast between predicate focus and sentence focus is clear in (14), from
Demuth (1989).
(14) Setswana
a. Monna o-fihl-il-e.
man 3sg-arrive-pfv-mood
‘The/*a man arrived.’
b. Ho-filh-il-e monna.
loc-arrive-pfv-mood man
‘There arrived a man,’ or ‘A man arrived.’
The basic form has default interpretations, both semantic and pragmatic.
The default interpretation is that the transfer of the flower to Yenghi was
completed, but this is not the only possible reading; it is also possible to say
without contradiction ‘but she did not receive it’. In terms of IS, the
unmarked focus position in Korean, as in other SOV languages, is the
immediate preverbal position (Kim 1988), and there are two ways to inter-
pret the AFD in predicate focus in (17a): it may not include Yenghi-eykey
‘Yenghi-dat’ (minimal AFD), or it may include it (extended AFD). In (17b)
the accusative case has replaced the dative on the recipient Yenghi; it has
‘spread’ from the undergoer to the non-macrorole recipient. When the
recipient is in the accusative case, it is impossible to add ‘but she did not
receive it’; it is a contradiction, because the accusative case signals that the
transfer was successfully completed. The accusative case canonically marks
the undergoer in Korean, and the undergoer codes the most affected partici-
pant in the event, as we saw in (9). Assigning the recipient accusative case
signals that its referent is affected by the action, which in the context of a
transfer event means that it got the item transferred.9
In (17c) the accusative case marker has been added to the dative-
marked recipient Yenghi-eykey, resulting in an RP with two case markers.
What does the accusative signal? It does not have the same
12.5.4 Step 4
The next step in the linking is selection of the syntactic template for the
utterance. The basic principles were introduced in Chapter 1, and they
made primary reference to the number of specified arguments in the LSs in
the semantic representation. The only reference to IS is indirect, mention-
ing displacement of arguments to the PrCS/PoCS (post-core slot) and the
occurrence of an argument-modulation voice construction, both of which
can be motivated by IS factors. The basic core syntactic templates, as
introduced in Chapter 1, are maximally unspecified and would not contain
any IS features. While this would be true for languages like English,
German, and all SOV and VSO languages, where the PFD includes the
whole clause, for SVO languages like French, Italian, Setswana and Seso-
tho, in which the PFD is constrained and excludes the core-initial, pre-
nuclear position, this could be indicated in the syntactic templates, since it
is a general constraint in the grammar of these languages. This is shown in
Figure 12.14, which represents the constraint that a pre-nuclear core argu-
ment is necessarily outside of the PFD. This constraint would be a feature
of all core templates in the language.
Figure 12.15 Syntactic templates for English wh-question and locative-inversion constructions
There are syntactic templates that are specialized for specific IS properties,
and they would have the AFD specified on them, since it is a feature of the
construction. Examples from English include narrow focus on a wh-XP in the
PrCS and the locative inversion construction (e.g. Into the room walked a tall
stranger), which is always sentence focus. They are illustrated in Figure 12.15,
from Van Valin (2005).
In a language with a PrCS which always expresses narrow focus, regardless
of whether the XP in it is a wh-expression or a plain XP, then there would be
no [þWH] feature, and the extent of the PFD would be left unspecified.
Italian is interesting, in that it combines the restriction in Figure 12.14 with
the wh-template in Figure 12.15, yielding a two-part PFD which includes the
PrCS, on the one hand, and the nucleus and post-nuclear constituents, on
the other, but excludes the pre-nuclear core-initial position (Bentley 2008).
This can be seen in Figure 12.16, which is the structure of (38b) in Chapter 11.
It means ‘A bike George bought, not a car.’ (The grey shading indicates the
discontinuous PFD.)
The PoDP RP la macchina ‘a/the car’ is outside the scope of the illocutionary
force operator of the first clause, due to its being in a detached position. It is
therefore necessarily part of the presupposed ICG.
12.5.5 Step 5
The final step involves mapping the elements in the sentence, as developed in
steps 1–3, into the syntactic templates determined in step 4, to yield the
actual form of the sentence, morphophonological processes aside. A big part
of this is linearization of the constituents in the sentence, something that IS
has at least some influence on in every language and a great deal of influence
in many. Numerous examples of the influence on the linearization of con-
stituents, including displacement to the PrCS, for example, are discussed in
the earlier chapter on IS. No further examples need to be introduced here.
One very important area of the interaction of IS and linking concerns
extraction phenomena and constraints thereon, which are discussed by
Shimojo in Chapter 16. IS plays a central role in explaining constraints on
extraction phenomena, and this not only makes possible a more principled
and motivated account of cross-linguistic variation in this domain, but it
also forms the basis of an empirically testable account of the acquisition of
these constraints by children. See Van Valin (1994, 1998, 2008) and Van Valin
and LaPolla (1997, Epilog).
(19) a. Lucy didn’t buy a new car. Sally did. Marked narrow focus
b. Lucy didn’t buy a new car. She leased one. Marked narrow focus
c. Lucy didn’t buy a new car. She bought a used Unmarked narrow focus
pickup.
d. Lucy didn’t buy a new car. She leased a new pickup. Predicate focus
e. Lucy didn’t buy a new car. Nothing happened. Sentence focus
getting the reading in (c) than in (b). Why should this be the case?
A comparison of (a) and (c) suggests the following: (c) is difficult because it
combines the marked (non-default) voice (passive) with marked narrow focus
on the PSA, a doubly marked sentence, which yields the same interpretation
as the maximally unmarked form, (a). In processing terms, the hearer must
process two non-basic patterns to arrive at the same interpretation that the
two basic patterns would yield, and it is difficult to justify the processing load.
Hence the preference for (a) over (b) with (i) is weaker than the preference for
(d) over (c) with (ii). This is an interesting interaction of IS with syntax, namely
PSA and voice selection, with semantic consequences.
The final area is intrasentential pronominalization. Following the pion-
eering work on IS and pronoun resolution by Kuno (1972a, b, 1975), Bick-
erton (1975) and Bolinger (1979), Van Valin (1990) and Van Valin and LaPolla
(1997: §5.6) proposed an analysis of sentence-internal pronominalization
based on the following principle.
Both of these conditions involve IS. The first one is illustrated in (21).
In (21a) Sam is the topic expression, and his sister is in the AFD, and accord-
ingly, Sam can be interpreted as being coreferential with his. In (b), on the
other hand, Sam is in the AFD and therefore cannot be interpreted as
coreferential with his. The second principle in many instances involves a
contrast between the pre-detached position (PrDP) and the PrCS, and the
phrases in them differ in terms of their IS function.
containing a possessive RP, but the PP is in the PrCS, not the PrDP, and this
means the possessor in the RP is in the AFD. Consequently, (b) violates
condition (20a), and coreference is impossible. In (b′), on the other hand,
condition (20a) is met, because the AFD is in the PrCS, and condition (20b)
does not apply because the pronoun is not in an argument position.
These three phenomena all involve IS and the interpretation of sentences,
which is consonant with the idea that the point of IS is to facilitate the
assimilation of information.
12.7 Conclusion
RRG began as a theory of grammatical relations back in the late 1970s, and
one of the fundamental claims made then was that the interaction of
discourse-pragmatics and semantic roles in the constitution of grammatical
relations is variably grammaticalized across languages, and these differences
in grammaticalization underly important typological differences among
languages. While in the ensuing decades the scope of the investigations
carried out in RRG expanded to more and more phenomena, the fundamen-
tal insight that grammatical systems crucially involve discourse-pragmatics
continues to be at the heart of RRG analyses, and this chapter, along with
those by Bentley, LaPolla and Shimojo, as well as the grammatical sketches
in Part V, illustrate this. The idea that discourse-pragmatics permeates gram-
mar is summed up in Figure 12.17, from Van Valin (2005: 182).
References
Balogh, Kata. 2021. Additive particle uses in Hungarian: A Role and Refer-
ence Grammar account. Studies in Language 45: 428–469.
Bentley, Delia. 2008. The interplay of focus structure and syntax: Evidence
from two sister languages. In Van Valin (ed.), 263–284.
Bentley, Delia. 2018. Grammaticalization of subject agreement on evidence
from Italo–Romance. Linguistics 56: 1245–1301.
Berio Leda, Anja Latrouite, Robert D. Van Valin, Jr. and Gottfried Vosgerau.
2017. Immediate and general common ground. In Patrick Brézillon, Roy
Turner and Carlo Penco (eds.), Modeling and Using Context, 1–14. Heidelberg:
Springer International Publishing.
Bickerton, Derek. 1975. Some assertions about presuppositions and pro-
nominalization. Communication and Linguistics Studies 11, Parasession on
Functionalism, 580–609.
Bolinger, Dwight. 1979. Pronouns in discourse. In T. Givón (ed.), Syntax and
Semantics, Vol. XII: Discourse and Syntax, 289–310. New York: Academic
Press.
Bresnan, Joan and Sam A. Mchombo. 1987. Topic, pronoun and agreement in
ˇ
Chichewa. Language 63: 741–782.
Demuth, Katherine. 1989. Maturation and the acquisition of the Sesotho
passive. Language 65: 56–80.
Demuth, Katherine and Mark Johnson.1989. Interaction between discourse
functions and agreement in Setswana. Journal of African Languages and
Linguistics 11: 21–35.
Filchenko, Andrey. 2007. Aspect [sic] of the Grammar of Eastern Khanty. PhD
dissertation, Tomsk State Pedagogical University.
Gundel, Jeanette. 1988. Universals of topic-comment structure. In M. Ham-
mond, E. Moravcsik and J. Wirth (eds.), Studies in Linguistic Typology,
209–239. Amsterdam: John Benjamins.
Gundel, Jeanette. 2012. Pragmatics and information structure. In Keith Allan
and Kasia Jaszczolt (eds.), The Cambridge Handbook of Pragmatics, 585–598.
Cambridge: Cambridge University Press.
Han, Jeonghan. 1999. On Grammatical Coding of Information Structure in Korean:
A Role and Reference Grammar Account. PhD dissertation, University at Buf-
falo (SUNY) [available on RRG website, https://rrg.caset.buffalo.edu/].
Kamp, Hans and Uwe Reyle. 1993. From Discourse to Logic. Hingham, MA:
Kluwer.
Kibrik, Alexander E. 1979. Canonical ergativity and Daghestan languages. In
Frans Planck (ed.), Ergativity, 61–78. London: Academic Press.
Kihara, C. Patrick. 2017. Aspects of G~ıkuy
~ u~ (Kikuyu) Complex Sentences: A Role and
Krifka, Manfred and Renate Musan (eds.). 2012. The Expression of Information
Structure. (The Expression of Cognitive Categories 5). Berlin: Mouton de
Gruyter.
Kuno, Susumu. 1972a. Functional Sentence Perspective: A case study from
Japanese and English. Linguistic Inquiry 3: 269–320.
Kuno, Susumu. 1972b. Pronominalization, reflexivization, and direct dis-
course. Linguistic Inquiry 3: 161–196.
Kuno, Susumu. 1975. Three perspectives in the functional approach to
syntax. CLS Parasession on Functionalism, 276–336.
Kuno, Susumu. 1991. Remarks on quantifier scope. In H. Nakajima (ed.),
Current English Linguistics in Japan, 261–287. Berlin: Mouton de Gruyter.
Kuno, S., K. Takami and Y. Wu. 1999. Quantifier scope in English, Chinese
and Japanese. Language 75: 63–111.
Lambrecht, Knud. 1987. Sentence focus, information structure, and the
thetic-categorial distinction. Berkeley Linguistic Society 13: 366–382.
Lambrecht, Knud. 1994. Information Structure and Sentence Form. Cambridge:
Cambridge University Press.
Latrouite, Anja. 2011. Case and Voice in Tagalog. PhD dissertation, Heinrich
Heine University Düsseldorf [available on RRG website, https://rrg.caset.
buffalo.edu/].
Latrouite, Anja. 2021. Specification predication: Unexpectedness and cleft
constructions in Tagalog. Faits de Langues 52(1): 227–254. doi: https://doi
.org/10.1163/19589514–05201011.
Latrouite, Anja and Arndt Riester. 2018. The role of information structure
for morphosyntactic choices in Tagalog. In Sonja Riesberg and Asako
Shiohara (eds.), Perspectives on Information Structure in Austronesian Lan-
guages. Studies in Diversity Linguistics, 247–284. Berlin: Language Sci-
ence Press.
Latrouite, Anja and Robert D. Van Valin, Jr. 2021. An RRG account of aspects
of the information structure-syntax interface in Tagalog. In Van Valin
(ed.), 257–283.
Levelt, Willem. 1989. Speaking. Cambridge, MA: MIT Press.
Nuhn, Patrick. 2021. Ay-Inversion in Tagalog: Information Structure and Morpho-
syntax of an Austronesian Language. Düsseldorf: Düsseldorf University Press.
O’Connor, Rob. 2008. A prosodic projection for Role and Reference Gram-
mar. In Van Valin (ed.), 227–244.
Park, Ki-seong. 1995. The Semantics and Pragmatics of Case Marking in Korean:
A Role and Reference Grammar Account. PhD dissertation, University at Buf-
falo (SUNY) [available on RRG website, https://rrg.caset.buffalo.edu/].
Planck, Frans. 1995. Double Case: Agreement by Suffixaufnahme. New York:
Oxford University Press.
Schachter, Paul. 1976. The subject in Philippine languages: Actor, topic,
actor-topic, or none of the above. In Charles Li (ed.), Subject and Topic,
491–518. New York: Academic Press.
Schachter, Paul and Fe Otanes. 1972. Tagalog Reference Grammar. Berkeley, CA:
University of California Press.
Schieffelin, Bambi. 1985. The acquisition of Kaluli. In Dan I. Slobin (ed.), The
Cross-Linguistic Study of Language Acquisition, 525–593. Hillsdale, NJ: Law-
rence Erlbaum.
Sgall, Petr, Eva Hajicova and Jarmila Panevova. 1986. The Meaning of the
Sentence in Its Semantic and Pragmatic Aspects, ed. by Jacob L. Mey. Dordrecht:
D. Reidel.
Shimojo, Mitsuaki. 2008. How missing is the missing verb? The verb-less
numeral quantifier construction in Japanese. In Van Valin (ed.),
285–304.
Van Hooste, Koen. 2018. Instruments and Related Concepts at the Syntax–Semantics
Interface. Düsseldorf: Düsseldorf University Press.
Van Hooste, Koen. 2021. A cross-linguistic survey of the instrument-subject
alternation. In Van Valin (ed.), 169–188.
Van Valin, Robert D., Jr. 1990. Functionalism, anaphora and syntax. (Review
article on S. Kuno, Functional Syntax.) Studies in Language 14: 169–219.
Van Valin, Robert D., Jr. 1994. Extraction restrictions, competing theories
and the argument from the poverty of the stimulus. In Susan Lima,
Roberta L. Corrigan and Gregory K. Iverson (eds.), The Reality of Linguistic
Rules, 243–259. Amsterdam: John Benjamins.
Van Valin, Robert D., Jr. 1998. The acquisition of wh-questions and the
mechanisms of language acquisition. In Michael Tomasello (ed.), The
New Psychology of Language: Cognitive and Functional Approaches to Language
Structure, 221–249. Mahwah, NJ: Lawrence Erlbaum [available on RRG
website, https://rrg.caset.buffalo.edu/].
Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cam-
bridge: Cambridge University Press.
Van Valin, Robert D., Jr. 2008. Some remarks on Universal Grammar. In J.
Guo, E. Lieven, S. Ervin-Tripp, N. Budwig, S. őzçalişkan and K. Nakamura
(eds.), Crosslinguistic Approaches to the Psychology of Language: Research in the
Tradition of Dan Isaac Slobin, 311–320. New York: Psychology Press.
Van Valin, Robert D., Jr. (ed.). 2008. Investigations of the Syntax–Semantics–
Pragmatics Interface. Amsterdam: John Benjamins.
Van Valin, Robert D., Jr. 2009. Privileged syntactic arguments, pivots and
controllers. In Lilián Guerrero, Sergio Ibáñez Cerda and Valeria A. Belloro
(eds.), Studies in Role and Reference Grammar, 45–68. Mexico: UNAM.
Van Valin, Robert D., Jr. 2016. An overview of information structure in three
Amazonian languages. In M. M. Jocelyn Fernandez-Vest and Robert D. Van
Valin, Jr. (eds.), Information Structuring of Spoken Language from a Crosslinguis-
tic Perspective, 77–92. Berlin: Mouton de Gruyter.
Van Valin, Robert D., Jr. (ed.). 2021. Challenges at the Syntax–Semantics–
Pragmatics Interface: A Role and Reference Grammar Perspective. Newcastle upon
Tyne: Cambridge Scholars Publishing.
Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning
and Function. Cambridge: Cambridge University Press.
von Heusinger, Klaus. 1999. Intonation and Information Structure. Habilita-
tionsschrift, University of Konstanz.
Notes
1 The background refers to the non-focal part of the utterance, namely the
two IUs outside of the AFD in Figure 12.2. It may or may not be part of the
pragmatic presupposition, although it usually is.
2 Focus–background and topic–comment represent two logically distinct
types of givenness: relational givenness in the former and referential
givenness in the latter (Gundel 2012).
3 The propositional representations are simplified for the sake of the
discussion; in a more technical presentation they would be represented
by the logical structures introduced in Chapters 1, 3 and 4.
4 For a pro-less analysis of the Mandarin phenomena discussed in Van Valin
and LaPolla (1997) mentioned earlier, see Van Valin (2005: 174–175).
5 Japanese verbs of giving and receiving are very complex with respect to
the social relationships and social statuses of the giver and the receiver.
None of that is relevant to this example, and therefore the verb of
receiving will be represented by an abstract ‘GET’ predicate.
6 See Van Valin (2009) for discussion of two languages (Liangshan Nuosu
[Lolo-Burmese] and Barai [Papuan]) in which IS plays an important role in
PSA selection, despite the lack of formal voice constructions in them.
7 This construction is often referred to as the ‘instrument subject alterna-
tion’. It is not possible in all languages, and in languages which have it, it
is highly constrained. See Van Hooste (2018, 2021) for a detailed discus-
sion of this construction.
8 See Nuhn (2021) for an in-depth RRG analysis of ay-inversion.
9 With intransitive verbs of motion, e.g. kan- ‘go’ as in Chelswu-ka san-ey kan-
ess-ta [Ch-nom mountain-loc go-pst-dec] ‘Chelswu went to(wards) the
mountain’, assigning the goal argument locative case means that
the moving participant may or may not have made it to the goal, as
the translation indicates. It is possible, however, to replace the locative
with the accusative case, Chelswu-ka san-lul kan-ess-ta, in which case the
sentence means that the moving participant did accomplish reaching
the goal, i.e. ‘Chelswu went to(*wards) the mountain’. Obviously, this
replacement of the locative by the accusative cannot be considered case
‘spreading’, since there is no other accusative RP in the clause. Rather, it
is better to speak of ‘case substitution’, which applies equally to this
example and to (17b).
13.1 Introduction
Most, if not all, grammatical theories have some framework for handling
complex sentences. The RRG theory of complex sentences, a.k.a. the
theory of clause linkage, was first systematically laid out in Foley and
Van Valin (1984, henceforth FVV), building on Olson’s work (1981). Since
then, significant revisions have been made, which one may find in Van
Valin and LaPolla (1997, henceforth VVLP) and in Van Valin (2005). The
RRG theory of complex sentences is unique in that, on the one hand, it is
based on the layered representation of clause structure, and, on the other
hand, it goes beyond the traditional coordination–subordination dichot-
omy, so that the theory provides a comprehensive framework which
captures both typological diversity and various facets of diachronic devel-
opment. In this chapter, we review the RRG theory of complex sentences
drawing upon a wide variety of languages. Due attention will be paid to
some of the recent developments both inside and outside the RRG
framework.
In the remainder of this section, the RRG theory of sentence structure
will briefly be reviewed as a backdrop for the later discussion. Section 13.2
will discuss the structural aspects of clause linkage, and Section 13.3 its
semantic aspects. Section 13.4 will provide a sample analysis of complex
sentences within RRG, focusing on the interaction of linkage types, oper-
ators and reference tracking. In Section 13.5 we offer some concluding
remarks.
The layered model of the clause is drastically different from the model of
clause structure in many other grammatical theories, which posit the
phrasal category VP. In RRG, there is no room for this category. This way
(2) Maori (Austronesian; Bauer 1993: 125. ‘>’ indicates the possessor–possessee
relation )
Me haere atu (raanei) ia ki toou kaainga, me haere
aux move away or 3sg to 2sg.gen>sg home aux move
mai raanei koe ki konei
hither or 2sg to here
‘He could come to your house or you could come here.’
Languages may vary with respect to the morphological strategy they adopt
for marking coordination. While all coordination markers in (2)–(4) (raanei
in (2), nee in (3) and táa in (4)) are words, a clitic or a coordinating particle
may also be used. Zero-marking is also not infrequent. In the Maori sentence
(2), raanei can be omitted, and in that case the resulting sentence would
allow either a conjunctive or a disjunctive interpretation, depending on
contextual factors. Generally speaking, the coordinating linkage is of a
‘balanced’ type (Croft 2001) and neither of the linked units is formally
reduced or de-ranked. Admittedly, coordinate constructions may have an
asymmetric interpretation such as temporal sequence and cause–result, but
this is usually taken to be an outcome of inference based on frame-semantic
knowledge and contextual information. Schematically, clause coordination
is represented in Figure 13.2. Here, there is no obligatory sharing of an
argument, and the linked clauses may describe separate events, each with
its own Actor-Undergoer assignment.
Next, core juncture is the linking of cores, that is, the predicate plus
obligatory arguments, and it involves partial sharing of argument
structure. The shared argument acts as something like a ‘hub’ for joining
two units.
In this example, the linked nuclei, helo’a ako ‘cook-do.for’ act like a simple,
unitary predicate, and all arguments are shared. The nucleus-level
coordination may be represented as in Figure 13.4.
The formation of a unitary predicate is a prominent characteristic of the
nucleus juncture. Compare (8) with (9).
While the same predicates occur in (8) and (9), the linkage structures are
different. In the former, two core arguments (marked by the non-nominative
core article te) follow the unitary predicate which acts like a ditransitive verb.
In (9), which is an instance of core juncture, each predicate takes its own
Undergoer core argument while sharing the Actor. In the linguistic literature,
the term serial verb construction (SVC) is often used inconsistently, but the
RRG typology of complex structures enables us to distinguish explicitly
between the linkage at the core level and that at the nucleus level.
In addition to the layered structure of the clause, RRG has other levels of
representation, namely the operator projection, information structure,
and logical structure. These systems interact with clause linkage in various
ways. Here, we shall look at the projection of operators in complex struc-
tures. According to Van Valin (2005: 9), the following operators can be
identified.
(10) Nucleus:
Aspect
Negation
Directionals (only those modifying orientation of action or event without
reference to participants)
Core:
Directionals (only those expressing the orientation or motion of one
participant with reference to another participant or to the speaker)
Event quantification
Modality (root modals)
Internal negation
Clause:
Status (epistemic modals, external negation)
Tense
Evidentials
Illocutionary force
When they occur in a coordinate linkage, operators that belong to the layer
higher than the layer where the linkage occurs necessarily take both linked
units within their scope. For example, clausal operators such as tense are
always shared by the linked cores (and, naturally, by the linked nuclei, as
can be seen in (8) and (9)). Core operators, for example root modality, are
always shared by the linked nuclei. An illustration of this interaction in the
Turkish example (7) is given in Figure 13.5. In linkages other than coordin-
ation, different situations hold, which we will discuss in the next section.
There are two other types of clausal subordination. A clause may occur in
an adverbial periphery modifying a clause, as in (12) (i.e. clause-to-periphery
or ad-clausal subordination), or it may serve as a constituent modifying a
noun, as in (13) (i.e. clause-to-noun subordination) forming a relative clause.
Here, the constituent Amy’s losing the race is embedded in the argument position
of another core, David regretted. The subordinated core is in participial form and
thus devoid of tense marking, which is a clause-level operator.
Next, example (15) illustrates nucleus-to-nucleus subordination, where the
second nucleus tέ ‘stay’ adds an aspectual meaning (namely progressive) to
the first nucleus gàrá ‘write’.
Since the nucleus does not have an argument position by definition, the
subordinate relation is that of adjunction, not embedding. In (15), the
second nucleus tέ ‘stay’ can be seen as an operator by itself, given its
aspectual meaning. This type of nuclear juncture is dubbed ‘operator con-
struction’ by Hasegawa (1996) for analogous cases in Japanese.
The constituent structures of core subordination (core-to-core embedding)
and nuclear subordination (nucleus-to-nucleus adjoining) are shown in Fig-
ures 13.9 and 13.10 respectively.
Here, the nexus type is distinguished from both coordination and subordin-
ation by the following properties. First, the first clause, bi-lisi-ni ‘(there) is
time’ is devoid of clause-level operator marking, while the second clause
aisi-zeŋe-i ‘I will mend the boat’ is marked for tense (namely future). The
form which indicates linkage, namely lisi, is a subtype of non-finite ending,
that is, it belongs to a conditional converb class. Second, the first clause is
not governed by any overt subordinating marker, nor does the linkage
involve embedding as in the case of subordination. Linkage like that in
(16) often results in a relatively long clause chain, a characteristic shared
by coordination but not by subordination. Third, the non-finite verb form
indicates the change of subject (notice that the first clause has the syntactic
subject eke ‘time’ and the second clause first-person singular), which is not
an available morphological option for the finite clause in this language.
In order to capture this type of nexus relation, that is, linkage involving
operator dependency but not embedding, RRG adopts the term
cosubordination, after the work by Olson (1981) on Barai. In addition to
these general properties, Roberts (1988) argues, based on evidence from
Amele, that cosubordination is differentiated from subordination on the
following grounds: (i) a subordinate clause can occur either initially or
finally with respect to the main clause, which is not possible for a
cosubordinate clause; (ii) the way pronominal reference works is different
in cosubordinate and subordinate clauses; and (iii) co-occurrence restric-
tions on conjunctive particles are different between subordinate and
cosubordinate clauses.
Admittedly, cosubordination is a highly schematic category and languages
may exhibit different concomitant properties within this broad definition.
The schematic structure of clausal cosubordination (16) is represented in
Figure 13.11. The operator projection is also given in order to visualize
clause-level operator dependency. It is true that not all cosubordinate con-
structions allow clause chaining and switch-reference marking, though they
are common among Papuan languages. However, the introduction of the
third type of nexus relation enables us to analyse clause linkage phenomena
in a more comprehensive and precise manner.
Cosubordination on the core layer is exemplified by (17). This is known as
a control construction, where the ‘missing subject’ of try is linked to the
‘higher subject’ John.
This sentence entails that John must wash the car. Hence the second core
wash the car is operator-dependent on the first core for deontic modality
expressed by must (VVLP: 460). Compare (17) with core coordination, exempli-
fied by (18).
From this sentence, it does not follow that Bill must wash the car, that is,
the second core wash the car is not operator-dependent on the first core. In
other words, the root modality operator must has scope only over the
first core.
Finally, in nuclear cosubordination, two linked nuclei form a unitary
predicate like nuclear coordination. Compare (19) with its core-level
counterpart (5). Unlike in core juncture, where the causee Jean occurs as the
‘hub’ for both linked cores, here this argument is expressed as an oblique
constituent marked by the preposition à (see also the Tukang Besi examples
(8) and (9)).
(19) French
Je ferai manger les gateaux à Jean.
I will.make eat the cakes to Jean
‘I will make Jean eat the cakes.’
In this example, two nuclei, namely korogari- ‘roll’ and oti- ‘fall’, are linked.
The aspect marker -tei- occurs next to the second unit, but it modifies both
nuclei. (20) entails that the economy is (metaphorically) rolling-and-falling
Coordination does not involve either kind of dependency. When two pre-
dicative constituents are coordinated, either can stand by itself. Subordin-
ation involves distributional dependency only; a subordinate constituent
can take its own operator that applies to the layer of linkage. Cosubordina-
tion involves both types of dependency, and, in this sense, it can be seen as
the most tightly integrated nexus relation. It is generally known that subor-
dination can be divided into two types, namely structures that involve
embedding in an argument position and those that do not (i.e. adjunction).
This matrix can therefore be further elaborated by introducing the feature
[argument] to capture these two subtypes of subordination. Since it is a
reasonable assumption that the argument position is more tightly inte-
grated into the clause structure than the adjunct position, we may arrange
the two subtypes of subordination in the following way:
Figure 13.15 Sentential subordination (parenthetical) in English (Kaltenböck et al. 2011: 849, orig.
from Burton-Roberts 2005: 180)
(24) The main point – why not have a seat? – is outlined in the middle paragraph.
The second recent revision in the typology of clause linkage concerns the
elaboration of the layered structure of the clause (see Bickel 1993; Van
Valin 2005). In addition to the initial formulation by FVV, RRG has accom-
modated the left-detached position (LDP), now called the pre-detached
position (PrDP), and the pre-core slot (PrCS) in order to capture the infor-
mation structure of the sentence. Let us see how they interact with clause
linkage in turn.
Generally speaking, the PrDP is reserved for providing a background
assumption, and is separated from the main clause by a prosodic break.
Here, the subordinate unit, moi, quand j’étais jeune ‘me, when I was young’,
has ‘its own left-detached expression, which makes it a sentence, and this
sentence is then in the left-detached position of the matrix sentence’ (Van
Valin 2005: 192). In this sense, (25) illustrates another type of sentence
subordination, namely sentence-to-PrDP subordination.
Next, the PrCS is distinguished from the PrDP in that it is a subpart of the
clause which is reserved for a marked or narrow focus and is not followed by
a prosodic break. The PrCS is typically occupied by an RP, but a clause may
also occur in this position (the sentence cannot occur in the PrCS by defin-
ition). Unlike in the case of sentential juncture, a clause which occurs in the
PrCS is devoid of illocutionary force. The contrast between sentence-to-PrDP
subordination and clause-to-PrCS subordination may be illustrated by the
following pair of sentences from Japanese.
(26) Japanese
Sainoo-ga ar-u-nda-kara ganbar-e.
talent-nom have-prs-cop-because work.hard-imp
‘Because/since (you) have a talent, work hard.’
(27) Japanese
Sainoo-ga ar-u-kara koso seekoo-si-tei-ru.
talent-nom have-prs-because foc success-do-res-prs
‘Precisely because/??since (X) has a talent (s/he) has made a success.’
In (26), the first predicative unit ends with the assertive predication marker
nda. The kara-marked unit expressing reason gives a presupposed piece of
information, and though the main clause is in the imperative form, this
presupposition remains unchallenged. The fact that the scope of the impera-
tive is limited to the second predicative unit supports the analysis of (26) as
an instance of sentence-to-PrDP subordination. In contrast, in (27) the link-
age marker kara is followed by the focus marker koso, indicating that sainoo-
ga ar-u ‘(X) has a talent’ occupies the PrCS, and hence (27) is an instance of
clause-to-PrCS subordination. Without the focus marker koso, a default
analysis of this sentence would be clause-to-periphery (¼ad-clausal) subor-
dination. As expected, inserting the assertive predication marker -nda before
kara in (27) results in a highly unnatural sentence because of the clash in
information structure, -nda strongly preferring to occur in the PrDP and koso
in the PrCS.
The structures of (26) and (27) would look like Figures 13.16 and 13.17.
So far, we have identified six types of subordination involving either a
clause or a sentence, namely clause-to-core (11), clause-to-periphery (¼ad-
clausal) (12), clause-to-noun (13), sentence-to-sentence (24), sentence-to-PrDP
(25)–(26), and clause-to-PrCS (27). In addition, Van Valin (2005) introduces
another kind of clause-to-periphery subordination, following the revision of
RRG to admit a periphery modifying the core, as in the following example
(Van Valin 2005: 194).
Here the clause after she arrived at the party is analysed as modifying the
core (as shown in Figure 3.18). A semantic motivation for analysing this
example in this way is that the specification of time is more relevant to the
propositional content of the clause compared to the semantic relations
expressed by markers such as although, because, and if, all of which have
more epistemic orientations. To keep this type of clause-to-periphery subor-
dination distinct from the structure like (12), the term ad-core
subordination is introduced. Examining to what extent this subdivision of
clause-to-periphery subordination is viable in the light of typological data
will be a topic for future investigation.
Besides the introduction of sentence-level junctures and the addition of
PrDP and PrCS as hosting positions for linkage, there have been debates on
the status of cosubordination. Most notably, Bickel (2010) and Foley (2010)
independently advance arguments against the notion of cosubordination.
Due to the limitation of space, we will not examine the validity of their
arguments here, but see Van Valin (2021) for a counter-argument.
The linkage types we have identified thus far can be listed as follows, in
order of stronger to weaker structural integration. For the sake of simplicity,
cosubordination is placed above subordination for each juncture type, but
the order between them is rather tentative compared to the relative order of
other relations.
(31) possessive
habitual actor
habitual agent
relative clause (making definite reference)
purposive complement
desire complement
indirect discourse complement
temporal adverbial clause
if–then
disjunction
conjunction
clause sequence (sequitur)
clause sequence (non-sequitur)
Silverstein’s idea was incorporated into FVV, and since then the list of
semantic relations, called interclausal semantic relations, has been updated.
Ohori (2001, 2005) proposed that the list of semantic relations should be
made more systematic in a way that derives certain relations from other
more primitive concepts. He introduced a feature system based on a small
number of conceptual distinctions, for example, [þ/ control], [þ/ action
coherence], and [þ/ temporal proximity], which capture stages of concep-
tual elaboration. This idea was partly adopted by Van Valin (2005: 206–207),
where the following list of semantic relations is proposed.
When the two hierarchies (30) and (32) are aligned, we obtain what is called
the Interclausal Relations Hierarchy (IRH). Naturally, languages differ with
respect to the types of commonly employed linkage types, and the range of
semantic relations that constructions of a given linkage type serve to express
also differs from language to language. Hence the typological generalization
RRG makes about the form–meaning correspondence in the realm of
(35) Japanese
Kare-wa kutibiru-o yugame-nagara warat-ta
He-top lip-acc twist-while smile-pst
‘He smiled while twisting his lips.’
(36) Japanese
Keehoo-ga nat-tei-nagara mina okunai-ni i-ta.
alarm-nom sound-prog-while everyone indoor-dat stay-pst
‘While the alarm kept ringing, everyone stayed indoors.’
In this sentence, nagara also connects two states of affairs that occur simul-
taneously. But unlike (35), the nagara-marked clause does not constitute a
subevent of the main clause, mina okunai-ni ita ‘everyone stayed indoors’.
What is more, (36) allows a concessive meaning besides simultaneity, as one
can imagine from the English gloss. Thus the relative strength of unit
integration on the structural side (namely core vs. clause cosubordination)
correlates with that on the semantic side (namely manner subevent vs.
concessive relation).
While the discussion in this section has been brief, evidence that will
reinforce the IRH can be garnered from a greater variety of languages.
More importantly, we may add a further dimension to the IRH by taking
into consideration more specific morphosyntactic features. In fact, one of
In this section, we shall see how the RRG theory of clause linkage serves
to enhance our understanding of specific grammatical phenomena,
focusing on Old Japanese (hence OJ) ‘switch-reference’ (Ohori 1992,
1994, 2002; see also McAuley 2002). It will be shown that what appears
to be a switch-reference system in OJ is in fact epiphenomenal, and the
apparent switch-reference function derives from the typology of clause
linkage. Put differently, reference tracking is a concomitant feature of
the clause-linkage structure, along with other important grammatical
features.
Let us start with Akiba (1977, 1978)’s study, where the uses of conjunctive
particles in the texts from tenth-century Japanese are closely examined.
Rejecting the idea that these particles should be characterized on semantic
grounds (for example, in terms of condition, cause, adversity, etc.), Akiba
asserts, ‘they are better characterizable in terms of the switch-reference
function’ (1977: 611). In fact, according to Akiba’s calculation based on
Taketori Monogatari, in 94 per cent of 536 occurrences of the linkage by te
the subject is retained across clauses, while in 92 per cent of 120 occurrences
of the linkage by ba there was switching of the subject. The following are
representative examples of linkage by te and ba. Because these markers are
semantically underspecified, they are glossed simply as link. Schematic
representations of the sentence structures are also provided.
Note that in both examples, the argument structure of the first linked unit
is intransitive and that of the second linked unit is transitive, and all
arguments are phonologically unrealized. Hence the only formal clue to
differentiate the working of reference tracking in these examples is the
choice of conjunctive particle. From these considerations, it appears to make
sense to analyse the two conjunctive particles, te and ba, as switch-reference
markers, the former coding the same-subject (SS) relation, and the latter the
different-subject (DS) relation.
However, a closer analysis reveals that te and ba differ in several respects
other than reference tracking, which cannot be explained if we assume that
switch-reference is the defining property of these particles. The primary
evidence for this interpretation comes from the interaction of operator
scope and the layer of linkage. In the RRG conception of sentence structure,
as we have seen in 13.2.1 (10), each layer of linkage has its own operator
specification. What is crucial about te and ba linkages is that they have
different possibilities for the occurrence of operators, in this case the
modal auxiliary.
First, consider the following example involving te-linkage. Here an epi-
stemic auxiliary meru occurs in the final unit of the linkage structure.
In (40), meru ‘may’ has scope over all the units linked by te, namely o-wo
sasage ‘raise (its) tail’, nana tabi meguri ‘go around seven times’ and umi-otosu
‘gives birth’. In contrast, in (41) the auxiliary besi has scope only over the
clause where it occurs, namely minami-no umi-ni huka-re-ohasi-nu-besi ‘(you)
would be blown to the south sea’. The difference between the two examples
can be represented schematically as follows, the underlined part indicating
the scope of modal auxiliary:
(40′) te-linkage
[unit 1]-te [unit 2] epistemic modality
(41′) ba-linkage
[unit 1]-ba [unit 2] epistemic modality
(42′) te-linkage
[unit 1]-te [unit 2] <-root modality
The behaviour of -mu as shown in (42′) is predictable from the RRG theory of
operators and clause linkage because root modality is an operator at the
core level and its scope does not extend to both linked units.
The next question concerning the reference-tracking properties of
clause linkage is why ba-linkage apparently has DS function. Unlike in
the case of te-linkage, where SS function derives straightforwardly from
the nature of clause linkage, the DS function of ba-linkage does not simply
follow from the fact that it is on the clause layer. Here the consideration
becomes largely a pragmatic one. Since ba-linkage is a clause-level con-
struction, it combines separate events. There, the ba-marked clause is
pragmatically detached, serving to provide the background assumption
‘given that/now that’, and this detachment provides a basis for the occur-
rence of a separate actor in the following clause. Also, because of this
detachment, the ba-marked clause assumes a separate viewpoint, which
accounts for the separateness of epistemic stance. Whether this pragmatic
status of the ba-marked clause should be reflected in the structural repre-
sentation, for example as a special kind of PrDP, is an interesting theoret-
ical issue, but at the moment, given the fact that the ba-marked clause is
capable of forming a clause chain, we shall remain rather conservative
about its status and not claim that it exemplifies a new type of
PrDP constituent.
To touch on the semantics of te- and ba-linkage, they both express tem-
poral sequence, but this similarity is only apparent. If we refine our defin-
ition of this notion so it involves temporal proximity, the difference
becomes clear: te-linkage expresses a closely connected sequence, which
often allows a manner or means interpretation, while ba-linkage is not
limited to immediate temporal sequence. This difference in temporal prox-
imity is also accompanied by different manifestation of causality. In te-
linkage, when there is room for causal interpretation, the relation is more
direct and the antecedent event enables the consequent event to happen,
while in ba-linkage, causality holds in the epistemic context, with the
antecedent event providing a reason for the statement given in the conse-
quent event. Here too, the relative strength of clause integration on the
semantic side exhibits correlation with that on the structural side, and this
correlation manifests itself in the difference of reference tracking across
linked units.
References
Abbreviations
14.1 Introduction
One of the significant features of Role and Reference Grammar (RRG) is that
it seeks to explain the universal features of clauses without imposing fea-
tures on languages in which there is no evidence for them.* A second
distinctive aspect is that it provides the necessary mechanisms for describ-
ing linguistic phenomena with scopes not only in simple clauses but also in
complex constructions. A third defining feature is the theory of clause
linkage. In RRG, the analysis of complex constructions is grounded in three
* This study was partially supported by CONACyT-Ciencia Básica (A1-S-24378) and the Ministerio de Ciencia, Innovación y
Universidades, Gobierno de España (FFI2017–85429-P) grants. I’m grateful to the anonymous reviewers and the editors of
this Handbook for their comments and suggestions on the first version of the chapter.
components: the level of juncture (nuclear, core, clausal and sentence), the
nexus relations (coordination, subordination and cosubordination), and the
semantic relationships between the two units. Unlike the traditional
coordination–subordination dichotomy, this model recognizes eleven
nexus–juncture combinations. The theory of clause linkage in RRG is crucial
to the analysis of adverbial (adjoined) clauses.
Typically, adverbial clauses are a type of subordination. When compared
with relative and complement subordinate clauses, however, adverbial
clauses are considered the least subordinate type (Matthiessen and Thomp-
son 1988; Givón 2001; Thompson et al. 2007). On one hand, adverbial
clauses do not fill an obligatory slot in the main clause and there is no
evidence that they are syntactically embedded (Diessel 2013); hence, adver-
bial clauses are usually considered adjuncts of the main unit in the periph-
ery or a peripheral position. On the other hand, they express several
semantic meanings including location, time, manner, purpose, condition,
reason, and so on; these meanings do not depend on the main verb or
clause. In fact, adverbial clauses subsume a wide range of structures with
varying syntactic properties that often overlap with the corresponding
features of coordinated constructions and other types of subordinated
sentences (Lehmann 1988; Cristofaro 2003; Hetterle 2015). For instance,
some but not all adverbial clauses are introduced by a subordinator or
clause-linkage marker (CLM); compare (1a) with (1b–d). Some CLMs have
lexical meaning (e.g. if in (1d)), while others do not (e.g. to in (1b)). Some
adverbial clauses show a flexible order with respect to the main clause (e.g.
temporal clauses (1c)), but others may prefer a fixed position (e.g. condition-
als (1d)). In addition, some adverbial clauses make use of tighter syntactic
linkages, while others are encoded by looser syntactic combinations.
A piece of evidence for a tighter syntactic linkage comes from the coding
of shared arguments: in (1a–b), the two units share the privileged syntactic
argument (PSA, i.e. traditional subject) and so it is coded only once in the
construction; in (1c), the PSAs are also the same, but each unit codes its own
argument. The marking of tense-aspect-mood (TAM) operators is also rele-
vant; in (1a–b), the linked verb depends on TAM operators from the main
verb, while in (1c–d), the linked verb is completely independent of TAM
marking and can be independently negated.
(1) English
a. Kim left yelling.
b. John went to the store to buy milk.
c. When he left the office, he forgot to turn the lights off.
d. If it rains, I won’t go to the movies.
The main goal of this chapter is to examine the set of rules that relates
syntactic and semantic representations to each other in adverbial (adjoined)
clauses.1 The analysis is based on argument sharing, operator sharing, and
the presence or absence of CLM, as well as the semantic cohesion between
the two units. In order to show the structural diversity of adverbial clauses,
the analysis is based on cross-linguistic data, though special attention is
given to English, Spanish and Yaqui, the languages I am most familiar with.
The content is organized as follows. Section 14.2 briefly introduces the RRG
theory of clause linkage, while Section 14.3 lays out the linking algorithm.
Section 14.4 examines the syntax and semantics of adverbial clauses; the
discussion is organized in five groups depending on the adverbial relation:
concessive and conditional clauses (14.4.1), reason clauses (14.4.2), temporal
clauses (14.4.3), purpose clauses (14.4.4), and manner and means clauses
(14.4.5). Section 14.5 summarizes the merits of RRG in capturing the com-
plex patterns of adverbial clauses.
RRG proposes three main components for the study of clause linkage: the
theory of juncture, the theory of nexus, and the interclausal semantic
relation. The first two parameters concern the syntactic representation of
complex constructions, while the last refers to their semantic representa-
tion. The theory of juncture deals with the units being linked. In a
nuclear juncture, there is a single core containing two nuclei; verbs in
a nuclear juncture take a single set of core arguments, that is, they form
a single complex predicate. In a core juncture, there is a single clause
containing more than one core, each with its own set of arguments; the
linked units can share part of the argument structure of the verbs. In a
clausal juncture, whole clauses are joined, and each clause may be fully
independent of the others. The schematic representation and examples
of juncture types are shown in (2). The examples come from Van Valin
(2005: 198).
arguments of the main verb, whereas the linked unit in (3e) is an adjunct
modifier of the main clause. Nuclear, core and clausal units modifying the
main clause are referred to as ad-nucleus, ad-core and ad-clausal subordin-
ation. The examples in (3), and part of the discussion that follows, come
from Matić et al. (2014: 5–9).2
temporal, and so on). Importantly, a given semantic relation may take more
than one juncture–nexus type. For instance, in Kokama (Tupian, Peruvian
Amazon), there are three purposive constructions that exhibit different
degrees of integration with the main unit. In the examples in (6), the linked
verb is not marked by TAM values, and the adverbial unit cannot take the
negative particles that operate in simple clauses (operator dependency); they
differ in terms of coreferentiality conditions (Vallejos 2014: 40). In the first
two clauses, there is a missing argument which must be coreferential with an
argument within the main clause; in tara constructions, the main absolutive
argument controls the obligatorily omitted nominative PSA (e.g. in order for
Magdalena to cure Jose), whereas in mira constructions, the main absolutive
controls the obligatorily omitted accusative argument in the linked unit (e.g.
in order for Jose to cure Magdalena). In tsen constructions, there are no
reference constraints, and all the purposive arguments must be expressed
by either RPs or pronouns; in fact, ra ‘3sg.m’ can refer to new entities (for
example, in order for someone else to cure Jose). Most likely, (6a–b) involve
clausal cosubordination, and (6c) clausal subordination.
(6) Kokama (Tupian; Vallejos 2014: 40)
a. Mijiri erura Mararina-uyi [Kutsi(¼pura) _i mutsanaka-tara]
Miguel bring Magdalena-pst1 Jose¼foc cure-clm
‘Miguel brought Magdalena in order for her to cure Jose.’
b. Mijiri erura Mararina-uyi [_i Kutsi(¼pura) mutsanaka-mira]
Miguel bring Magdalena- pst1 Jose¼foc cure-clm
‘Miguel brought Magdalena in order for Jose to cure (her).’
c. Mijiri erura Mararina-uy [Kutsi(*¼pura) ra mutsanaka-tsen]
Miguel bring Magdalena- pst1 Jose¼foc 3sg.m cure-clm
‘Miguel brought Magdalena in order for her/another to cure Jose.’
Strongest Closest
Nuclear co-subordination Causatives [1]
Phase
Nuclear subordination Manner
daughter Motion
peripheral Position
Nuclear coordination Means
Psych-action
Core co-subordination Purposive
Jussive
Core subordination Causatives [2]
daughter
peripheral
Direct perception
Core coordination Indirect perception
Propositional attitude
Clause co-subordination Cognition
Indirect discourse
Clause subordination Direct discourse
daughter Circumstances
peripheral Reason
Clause coordination Conditionals
Concessive
Sentential subordination Simultaneous actions
Sequential actions
Sentential coordination Situation-situation: unspecified
Weakest Loosest
Syntactic relations Semantic relations
Figure 14.1 Interclausal Relations Hierarchy (Van Valin 2005: 209)
are less tight than linkage types at the top end. In the middle portion, in a
non-subordinate core juncture, the two cores obligatorily share one core
argument; in a subordinate core juncture, the linked core serves as a
syntactic argument of the matrix core. The semantic relations are also
organized according to the degree of semantic cohesion between the two
units, that is, the degree to which they express facets of a single action or
discrete actions or events.
There is an iconic principle governing the interaction of the two hierarch-
ies: the closer the semantic relation between two propositions, the stronger
the syntactic link joining them. That is, the semantic relations at the top
end of the hierarchy should be realized by the linkage categories at the top
as well, and the relations at the bottom of the hierarchy should be realized
by the linkage categories at the bottom of the syntactic side. Moreover, while
there is often more than one syntactic realization of a particular semantic
relation, the tightest syntactic linkage realizing it should be tighter than the
tightest syntactic linkage realizing looser semantic relations. The syntactic
relations and the semantic relations are linked together following a set of
mechanisms or linking algorithm.
Within RRG, the algorithm system links the syntactic, semantic and prag-
matic components in a sentence, and it includes universal as well as
language-specific components (Van Valin 2005: 129). The system is bidirec-
tional, that is, it links the semantic representation to the syntactic repre-
sentation (language production), and the syntactic representation to the
semantic representation (language comprehension). The algorithm for
linking semantics to syntax for simple sentences is summarized in (8); see
also Chapter 1 of this handbook.4
a. If the verb is intransitive, then assign the PSA either to the macrorole
or direct core argument status, depending upon the language
(language-specific)
b. If the verb is transitive and the language lacks voice oppositions,
determine the macroroles from case marking and/or word order
(language-specific)
c. If the language has a voice opposition, determine the voice of a
transitive verb (language-specific)
1. If the construction is syntactically accusative:
a. if it is the unmarked voice, the PSA is the actor
b. if it is passive, the PSA is not the actor of the predicate in the nucleus
1. the actor may appear as a direct core argument (language-
specific); or
2. the actor may appear in the peripherycore marked by an
adposition or oblique case (language-specific); or
3. if there is no actor in the core or the periphery, then replace
the variable representing the highest ranking argument in
the LS with ‘Ø’.
2. If the construction is syntactically ergative:
a. if it is the unmarked voice, the privileged syntactic argument is
undergoer.
b. if it is antipassive, the privileged syntactic argument is actor:
1. the undergoer may appear as a direct core argument or as an
oblique element (language-specific);
2. if there is no undergoer in the core or the peripherycore ,
then replace the variable representing the lowest ranking
argument in the LS with ‘Ø’.
3. Assign macrorole status to the other direct core argument, if it is
not dative or in an oblique case (language-specific).
d. If the language is head-marking and there are independent RPs in the
clause, associate each RP with a bound argument marker (language-
specific).
2. Retrieve from the lexicon the LS of the predicate in the nucleus of the
clause and execute step 2 from (8) with respect to it, subject to the
following provisos:
a. If the language allows variable undergoer selection and if there is more
than one choice for undergoer, do not assign undergoer to an
argument in the LS.
b. Determine the linking of the non-macrorole core argument:
1. If there is a two-place state predicate in the LS and if the non-
macrorole core argument is marked by a locative adposition or
dative or a locative-type case, then link it with the first argument
position in the state predicate in the LS and link the other non-actor
core argument (if there is one) to the second argument position in
the state predicate, or
2. If there is a two-place state predicate in the LS and if the non-macrorole
core argument is not marked by a locative adposition or dative or a
locative-type case, then link it with the second argument position in the
state predicate and link the other non-actor core argument (if there is
one) to the first argument position in the state predicate.
Most complex sentences follow the linking system developed for simple
clauses. For instance, the units involving clausal junctures behave largely
like independent clauses. Consider the following English constructions (Van
Valin 2005: 228–229).
(10) a. Dana jogged through the park, and Kim waved to him
b. Kim worked on the assignment in the morning and __i/*j will finish it in
the afternoon
(11) Yaqui
a. Inepo teopo-u bwite-k into (ne) Peo-ta bicha-k
1sg.nom church-dir run.sg.pfv clm 1sg.nom Peter-acc see-pfv
‘I ran to the church, and then (I) saw Peter.’
b. [U-me chapayeca-mi yi’i-su-k-o], _i maska-ta emo u’ura,
the-pl chapayeca-pl dance-cmpl-pfv-clm mask-acc self take.off
teopo bicha-po _i im am¼jajawa, emo koba-t _i
church towards-loc here 3pl.acc¼red.leave self head-loc
payum-mea papatta, juya-u-bicha _i sasaka-k
cloth-ins.pl red.cover hill-dir-towards red.go.pl-pfv
‘After the Chapayecasi danced, they took off the masks, _i/*j put them in
front of the church, _i/*j covered their heads with cloths, and _i/*j went
towards the hill.’
In the analysis that follows, adverbial relations are organized into five
groups: concessive and conditional clauses (14.4.1), reason clauses (14.4.2),
temporal clauses (14.4.3), purpose clauses (14.4.4), and manner and means
clauses (14.4.5). Adverbial relations may be established at different levels of
the clause structure; the relationships at the bottom of the semantic hier-
archy usually modify the clause, temporal and purpose clauses may modify
the core, and manner and means can usually modify the nucleus.
hypothetically happen, as in If he gets the job, we’ll all celebrate, or that did not
happen, as in If you had been at the party, you would have seen her new boyfriend
(Hetterle 2015: 49). In the English constructions in (12), each clause links
separately, but the complete construction encodes two semantically asso-
ciated states of affairs. Each association is formally captured in the (sim-
plified) logical structure of the construction by different connectors or
CLMs with lexical content: IN . SPITE . OF ′ for concessive (12a) and for
conditional (12b).
(12) English concessive and conditional clauses (Van Valin 2005: 207)
a. Bill made it to work, even though it was snowing heavily.
a′. [Bill made it to work] IN . SPITE . OF ′ [it was snowing heavily]
b. If it rains, we won’t be able to have a picnic.
b′. [If it rains] [we won’t be able to have a picnic]
SENTENCE
PERIPHERY CLAUSE
CORE
CLM CLAUSE
CORE
Some languages have several types of reason clauses. For instance, Cavineña
distinguishes cause clauses from reason clauses (Guillaume 2008). The ¼ra
clause in (14a) expresses the cause of the main event, whereas the ¼tibu clause
in (14b) gives the reason for the occurrence of the main event. Unlike cause
clauses, the linked verb in reason clauses can take the full range of verbal
morphology, and there are no coreference restrictions.
‘although, even though’ (15a), and kiali’ikun ‘because, so that’ (15b). Second,
the description of the event in the adverbial unit is fully independent in
terms of TAM, negation and IF. Third, concessive and reason clauses can
attract restricted discourse markers that occur on the left edge in inde-
pendent clauses and can have their own pre- or post-detached phrases
(PrDP, PoDP). Finally, the dependent PSA takes nominative case; nomina-
tive PSAs are completely banned in syntactically subordinate clauses in
Yaqui (Guerrero 2006, 2017).
(15) Yaqui concessive and reason clauses
a. Bempo Guayma-meu saja-ne [ella’apo empo im tawa-ne]
1pl.nom Guaymas-dir.pl go.pl-pot clm 2sg.nom here stay-pot
‘We will go to Guaymas although you stay here.’
b. Inepo into in jaboii¼tua, tei tu’isi emo jala’eka-n,
1sg.nom and 1sg.gen grandpa¼int 1pl.nom good self friend.be-ipfv
[kiali’ikun ne au¼waate into ne a-et ai ¼ baisae-Ø
clm 1sg.nom 3sg.obl¼miss-prs and 1sg.nom 3sg-loc 3sg.acc¼thank-prs
si’ime-tai ]
everything-acc
‘As for me and my grandpa, we were very good friends, so I miss him and I thank him for
it, for everything.’ (Buitimea; regreso: 37)
Therefore, reason and concessive clauses in Yaqui link together two independ-
ent units in terms of an initial CLM, operator and argument coding; in (15b),
there are two complete sentences, each with its own detached position: there
is a topical element in the PrDP of the first component, as for me and my
TEXT
Inepo into in jaboii=tua, tei tu’isi emo jala’eka-n, kiali’ikun ne au=waate into ne a-et ai=baisae-Ø si’ime-tai
‘As for me and my grandpa, we were very good friends, so I miss him and I thank him for it,
for everything’ in (15b)
Figure 14.3 Simplified structure of reason relations as sentential coordination in Yaqui
grandpa, and there is an anti-topic element in the PoDP of the second unit,
everything (sentential coordination). In contrast, conditional clauses show two
key features signalling structural dependency: the presence of a final, bound
and underspecified CLM, and operator dependency; (16c) also satisfies the
need for an accusative dependent PSA (ad-clausal subordination).
A simplified syntactic representation of (15b) is offered in Figure 14.3.
In order to capture the semantic association of ad-core (14a) and ad-
clausal (12, 13, 14b, 16) adjuncts, Van Valin (2005: 229) revised step 4 in
the syntax-to-semantic algorithm in (9). The relevant modifications are in
boldface in (17). Step 4b allows us to represent concessive, conditional and
reason PPs or clauses in the periphery. Sentential coordination for the
Yaqui constructions in (15) does not require any modification to the
linking algorithm.
(17) Revision of step 4 in the syntax-to-semantics linking algorithm (Van Valin 2005:
229)
4. If there is an adjunct phrase in a periphery,
a. and if it is in the peripherycore , then retrieve the logical structure of the
predicative adposition from the lexicon, insert the logical structure of
the core as the second argument in the logical structure and the object
of the adposition as the first argument,
b. and if it is in the peripheryC L A U S E , then link the adjunct PP or clause
logical structure to the matrix logical structure via the semantic
representation of the predicative adposition or CLM.
than conditional and reason clauses. For instance, temporal relations (but
not reason relations) involve the predetermination of time and aspect values
of the linked state of affairs, and might often include reduced verb forms
(Cristofaro 2003: 173). Across languages, temporal relations can be expressed
by different syntactic means (adverbs, adpositional phrases, finite and non-
finite clauses), and can appear to the left or to the right of the core they
modify (i.e. as modifiers in the periphery of the core).
The semantic association between the two units can be captured by CLMs
with lexical content. For instance, ‘before’ expresses posterior relations with
respect to the main event, ‘after’ introduces anterior relations, and ‘while’
introduces co-temporal events. In Spanish, the prepositions antes (de) ‘before’
and después (de) ‘after’ can take phrases, cores or clauses as objects (Guerrero
et al. 2017). In (18a–c), there is a subjectless, non-finite infinitival unit. As
mentioned before, infinitival verbs are a type of non-subordinate core that
depends on the main verb in terms of TAM operator. However, the infinitival
unit in temporal clauses can be independently negated from the main
clause, as in Sam se durmió después de no cenar ‘Sam fell asleep after not having
dinner’. Commonly, the PSA of the main core (Sam) controls the referential
identity of the missing argument (pivot) within the linked core, but in (18c)
each core has its own PSA. This means that argument sharing is optional.
Hence, since the two cores can be independently negated, and they can but
do not necessarily share an argument, (18a–c) involve core coordination.
There is an additional sequential construction. In (18d), the linked unit is
introduced by the complex CLM después de que ‘after that’ and the two units
are independent of each other in terms of TAM operators at the level of the
clause and argument coding. Therefore, (18d) takes a modifying clause at
the periphery of the core, hence ad-core subordination. In Spanish, the
unmarked position for temporal adverbs, phrases and clauses is postverbal,
though they may also appear at the beginning, as in antes de acostarse, Sam
besó a Pat for (18a). The simplified logical structure for sequential construc-
tions is the same as that of a peripheral PP modifying a core and is valid for
the two juncture–nexus.
SENTENCE
CLAUSE
CORE PERIPHERY
PP
COREp
NUCp CLAUSE
Peripheral subordination at the core level does not require any revision
of the linking algorithm, as it is captured by the steps in (5d1a) in the
semantics-to-syntax linking (8): the predicative preposition takes either a
core or clausal unit and, as a whole, is attached to the periphery of the core.
When linking syntax to semantics, step 4a yields the representation for
spatial and temporal circumstances, based on the predicative adposition.
A simplified syntactic representation for the sequential construction in (18d)
is presented in Figure 14.4. This is another example of asymmetrical linkage
since a linked clause is contained within a sub-clausal unit.
Core coordination junctures do require revision of the linking algorithm,
especially when the main actor controls the pivot in the linked unit, as in
(18a–b). In order to capture the fact that there is a syntactic argument slot
missing in the linked unit, the template selection principles in the seman-
tics-to-syntax linking (8) were revised, and a new principle (step 4b) was
added for non-subordinate core junctures. Step (4b1) specifies that the
linked unit is a semantic argument of the matrix verb, as in Chris tried to
see Pat. However, temporal clauses do not fulfil a semantic slot in the logical
structure of the main clause. The nature of Spanish temporal infinitival
cores in (18a–b) can be captured by step (4b2).
The statement in step 4b does not specify which syntactic slot is missing,
since that is a construction-specific feature. The syntax-to-semantic linking
needs some modifications too. The relevant adjustments apply after the
logical structure of the clause has been retrieved (step 2), and the linking of
macrorole and non-macrorole core argument has taken place (step 3). Again,
step 4b can capture the fact that there is a syntactic argument missing from
the linked core which must be interpreted as being the same as one of the
syntactic arguments from other positions. The logical structures for (18a)
and (18b) contain a lexically unfilled co-indexed argument x, which is
controlled by the PSA of the main clause (Sam).
That is, all temporal clauses in (18) semantically modify the core of the main
unit, but only the linked unit in (18d) occupies a peripheral position; the
non-finite linked units in (18a–c) yield a non-subordinate core juncture.
A simplified representation for core coordination expressing sequential
relations in Spanish is presented in Figure 14.5.
Depending on their role in discourse, temporal clauses can also occupy
the pre- or post-detached position, with or without a prominent pause
between the two units (Guerrero et al. 2017). In a sentence like Ricardo
S fue expulsado en 1906, después de pasar dos años removiendo el caso de la muerte
de DG ‘Ricardo S was expelled in 1906, after spending two years investigating
the death of DG’, there is a non-finite core, and the missing argument is co-
indexed with the main actor. In this construction, there are two temporal
expressions, the PP ‘in 1906’, and the ad-clause introduced by the prepos-
ition después de ‘after’. Because there is another core modifier, I suggest that
the after-clause does not modify the core, but is linked to the sentence
directly in the post-detached position (PoDP). Notice that there is another
SENTENCE
CLAUSE
SENTENCE
CLAUSE PoDP
CORE PERIPHERY PP
PP CORE
COREp NUCp CORE
Ricardo S fue expulsado NUCp RP PRED
PRED p
p
en 1906, después de
pasar dos años removiendo el caso de la muerte de DG
‘Ricardo S was expelled in 1906, after spending two years investigating the death of DG’
may wonder what the logical structure of the constructions in (21) looks
like. The Interclausal Semantic Relation Hierarchy (Van Valin 2005: 206–207)
recognizes temporal circumstances as the spatial or temporal parameters of
an event (e.g. Kim saw Pat after she arrived at the party) and temporally ordered
states of affairs. There are simultaneous states of affairs, where one state of
affairs is temporally coterminous with another, like in (22a), as well as
sequential states of affairs, in which one state of affairs follows another
temporally, with or without any temporal overlap, as in (22b). Note that the
linkage markers do not have lexical meaning but resemble some sort of
conjunction/connector. That is, the semantic association between the units
cannot be captured by the CLMs, but instead depends on the pragmatic
context.
(22) a. Max was dancing, and at the same time Susan played the piano.
a′. [do′ (Max, [dance′ (Max)])] ^ [do′ (Susan, [play′ (Susan, piano)])]
b. Juan finished reading the newspaper, and then Carlos walked into the room.
b′. [do′ (Juan, [read′ (Juan, paper)])] & [do′ (Carlos, [walk′ (Carlos) & INGR be-at′
(room, Carlos)])]
In Cutrer’s analysis (1993: 177), purpose clauses are a type of core juncture,
and therefore, there must be an argument shared between the two cores. It
is possible to have two control relations, as in (23b), but only one is obliga-
tory: the postverbal gap must be obligatorily controlled by the main theme.
d′. [do′ (deer, [run′ (deer)] PURP [do′ (man, Ø) CAUSE [NOT BECOME dead′ (deer)]]]
e. Min-Øi yoi-taj kaba’i-mkreuwa-k [_j amk¼wiria-ne-betchi’ibo]
Fermín-nom foreigner-acc horse-pl lend-pfv 3pl.acc¼feed-pot-clm
‘Fermín lent the foreigner the horses in order for him to feed them.’
e′. [do′ (Fermín, [lent′ (Fermín, horses)] CAUSE [BECOME have′ (foreigner, horses)]] PURP
[do′ (foreigner, [feed′ (foreigner, horses)])]
Notice that Yaqui does not distinguish between purpose (25a–d) and
rationale clauses (25e) based on an obligatory syntactic gap, since a referen-
tial controlled undergoer must be overt. In the last example, the two cores
share both the theme and the recipient, which are lexically expressed as
core arguments of the main unit; in the linked unit, the recipient yoi
‘foreigner’ controls the actor PSA of the linked verb ‘feed’, hence there is
structural control (missing syntactic argument), while kaba’im ‘horses’ con-
trols the undergoer of ‘feed’, and so there is inherent control (co-index
pronoun). Since there are unfilled/filled syntactic arguments in the linked
core which need to be obligatorily controlled by a core argument in the
main unit, Yaqui purpose clauses behave as a kind of control construction
(Guerrero 2017).9 As a matter of fact, the special lexical coding of the
controllee is exactly the same in purpose clauses and constructions taking
actor and undergoer control verbs. In Yaqui, actor control verbs such as
modal, phasal and desiderative constructions demand a cover controllee in
the linked unit (26a), whereas undergoer control verbs like causative and
jussive constructions demand an overt controllee, as depicted in (26b). In
purpose, causative, and jussive constructions, a missing argument co-
indexed with the main theme would be ungrammatical.
(27) Comanche (Northern Uto-Aztecan; McDaniels 2014: 75, 73, 77, 83)
a. Yuhu-noko-pī-aj nīi tīmī-i [nīi _j tīhka-tui]
fat-bake-res-acc 1p(s.clitic) trade-pfv 1p eat-unr
‘I bought frybreadj in order for me to eat _ j’.
b. S-u-tī-sei peekwij kwīhī-n [uj _i tīhka-tui]
def-3p-nom-dm fish catch-pfv 3p(o.clitic) eat-unr
‘Hei caught a fish (in order) _ i to eat it’.
c. S-u-tī-sei u-hkupa uj tīīki [u-hkaj pīi napinai-tui]
def-3p-nom-dm 3p-inside 3p(o. clitic) put 3p-acc 3p.co save/select/stash-unr
‘Hei put it in there (in order) for himi to set it aside’.
d. S-u-tī-sei tī-poo-pī-aj [pīi _j niha-tui-a] yaa-tī
def-3p-nom-dm unsp.o-make.mark-res-acc 3p.co.gen read-unr-acc take-ipfv
‘He is taking a booki to read _i’.
b. If the matrix predicate is not a control verb, then link the unlinked
syntactic argument in the matrix core to the logical structure
argument position of the pivot of the linked core, otherwise
c. If the non-subordinate core juncture expresses a purpose relation, and
1. if the construction demands structural control, then the linked
argument must be unfilled (e.g. syntactic controllee in English,
Lakhota, Dyirbal); otherwise,
2. if the construction demands inherent control, then the linked
argument can be filled by a bound pronoun, clitic or agreement-
inflection, (e.g. semantic controllee in Acehnese, Chuj, Yaqui,
Toqabaquita)
3. the identity of the controllee may follow the referential control
hierarchy:
the referential identity of an obligatory controlled argument is
determined by the main actor > the main undergoer > another
main core argument > there is no control relation.
SENTENCE
CLAUSE
CORE
SENTENCE
CLAUSE
Table 14.2 Constructional schema for ‘I bought frybread in order for me to eat’ in
(27a)
In Yaqui, these adverbial relations are coded by the general CLM -kai. In the
constructions in (30), -kai introduces the manner in which a motion event is
carried out (30a), and the means by which an action is carried out (30b). The
example in (30c) expresses some sort of positive/negative circumstance
around the main action. The entire clause describes a complex event in
which the main actor is engaged. Because -kai is a general CLM, the actual
semantic relation between the two units is recovered from the discourse
context, the nature of the events, and world knowledge (e.g. two ongoing
events involving the actor; a functional answer for why). In the logical
representations, ^ represents two simultaneous states of affairs. Note that
there is an unfilled position in the second component co-indexed with the
main PSA.
The linked unit marked by -kai is not syntactically subordinated to the main
event, but adds a second predication to it. Several major features define
manner and means kai-clauses in Yaqui. The linked event can denote a state as
well as an activity predicative (manner is not limited to motion events, for
instance), and the two events are co-temporal (simultaneous). Additionally,
manner and means clauses are usually reduced units (intransitive verbs or
transitive verbs without referential core arguments), the verb must be
unmarked for TAM affixes and negation, and must generally occur at the
beginning or the end of the sentence without a pause. More importantly, they
are subjectless clauses. The main PSA (the actor) must control the identity of
the missing syntactic argument in the linked unit. In sum, this clause-linkage
type yields nuclear or core cosubordination depending on the valence of the
linked verb. In a nuclear juncture, the two nuclei combine to form a complex
nucleus with a single set of core arguments (30a–b). Figure 14.9 gives the
syntactic representation of nuclear cosubordination in (30b).10
In a core juncture, the two cores share one but not all arguments, as in
(30c) and (31b–c). Figure 14.10 gives the syntactic representation of core
cosubordination in (31c). The actual linking of this construction is the same
as the one in rationale/purpose clauses, in which there is a missing syntactic
argument in the linked unit co-indexed with the main PSA, and the linked
verb is unmarked for TAM suffixes.
SENTENCE
CLAUSE
CORE
RP NUCLEUS
NUC NUC CLM
SENTENCE
CLAUSE
CORE
PrCs
CORE CLM CORE
RP
14.5 Conclusion
The main goal of this chapter was to examine how the linking algorithm
works in adverbial (adjoined) clauses. I have argued that the structural
linkages which express adverbial relations across languages are heteroge-
neous: some relations make use of looser syntactic combinations, while
others are encoded by tighter syntactic linkages. Furthermore, the same
semantic relation can show more than one syntactic linkage with different
degrees of semantic and syntactic integration. The theory of clause union in
RRG, based on juncture–nexus relations, conveniently accounts for these
structural complexities. On one hand, adverbial relations can be coded by
ad-subordination, as well as those relations expressed by coordination and
cosubordination. On the other hand, only ad-subordinated linkages occupy a
peripheral position at the level of the core or the clause, whereas non-
subordinated linkages demand a different syntactic representation.
References
Notes
15.1 Introduction
Role and Reference Grammar (RRG) holds the view that a sentence is not just
a concatenation of words into a sequence but the incarnation of a set of
formal templates that are stored in syntax.* Syntactic templates have two
effects that are crucial for the understanding of the nature of linguistic
symbols. First, they are a key factor in determining the pattern followed by
the influx of information contributed by each word into the new symbol (i.e.
the phrase). Second, the phrase is an enclosure where certain relations occur
among its constituents that are shielded from some external influences.
Cleft sentences and relative clauses make this encapsulation of the phrase
evident. Sentence (1) illustrates a typical instance of a relative clause struc-
ture (hereafter, RC).
(1) The student who failed the test quit the program.
* This paper has benefited greatly from the careful reading, insightful questions and several suggestions by Delia Bentley, one
of the editors of this Handbook. Of course, I am responsible for all remaining errors.
relative clause (though see Section 15.2 for some relevant restrictions in
Malagasy, a Western Austronesian language spoken in Madagascar). Therefore,
the shared participant is not required to play any specific syntactic or seman-
tic role in either the matrix clause or the relative clause. This has important
consequences for the syntax of RCs. The relative clause is safeguarded from
any requirement or influence from the matrix clause by the RP in which it is
embedded. Operators like tense, grammatical aspect or negation affect each
clause (matrix and relative) independently. All the formal properties imposed
on the relative pronoun come from the head noun – and are of a referential
nature (person, gender and number) – or from the matrix verb of the relative
clause. These properties are captured by an analysis of relative clauses as RP
nuclear modifiers (Pavey 2004; Van Valin 2005: 220–223; 260–266; Van Valin
and LaPolla 1997: 497–499). This means that RCs, just like English attributive
adjectives, are elements in the periphery that describe either a property of the
referent or an event in which it participates. This semantic dependency is
reflected in the syntax and further justifies the nuclear periphery analysis: for
those languages with a rich nominal morphology, the noun imposes agree-
ment relations on the attributive element.
The semantics of RCs displays properties that illustrate the insufficiency of
truth conditional semantics to account for linguistic meaning. The meaning
of the relative clause and the head noun in (1) can be expressed by the
following simple sentence.
(2) The student failed the test (and quit the program).
Thus, the information about the world is the same in both cases. However,
the meanings are quite different because the composition of the respective
semantic representations is different. This will be captured later in this
chapter (see the representation in (10)) by embedding the logical structure
(LS) of the RC as an argument of the attributive LS, which is consistent with
the RC syntax, in that the relative clause is a modifier of the RP nucleus. In
addition to EHRCs, Section 15.2 will also discuss Internally Headed Relative
Clauses (IHRCs), free relatives and non-restrictive relatives. In IHRC, the head
noun is inside the embedded clause, which we know is embedded due to the
presence of a complementizer. The so-called ‘free relatives’ are not really
RCs, since they lack a head noun as well as a relative pronoun in many
languages. Finally, in non-restrictive RCs, the embedded clause is separated
from the head noun by a pause, and it does not fulfil the same pragmatic
function characteristic of relative clauses, namely, the identification of the
referent of the head noun.
This chapter also discusses cleft sentences, specifically, it-clefts as exempli-
fied by (3), which will be the focus of Section 15.3.
The components of it-clefts are a matrix clause with a ‘dummy’ ‘it’ pronoun,
a copula clause and a clefted noun phrase (John in (3)), which is the
antecedent that fixes the reference of either a relative pronoun or, if this is
absent, an argument or adjunct slot in the embedded clause, as in (4).
The ingredients of a typical English RC like (1), repeated in (6a) for ease of
exposition, are a clause introduced by a relative pronoun (i.e. who) co-indexed
with a head noun outside the RC and preceding it (i.e. student) and, finally, a
matrix clause, which the head noun belongs to. The RC is embedded within
the RP headed by the antecedent noun – student – as can be seen, for example,
by the fact that a proform can replace the entire RP (see 6b).
(6) a. The student who failed the test quit the program.
b. She quit the program.
The events described by the RC – the failing event – and by the matrix
clause – the quitting event – might not be connected beside the fact that
they share the same participant. A central relation within the RP is that of
obligatory co-indexation between the head noun and the relative pronoun.
As long as it belongs to the RP, there may be intervening material – all
within the RP that contains the relative clause – between the head noun and
the relative pronoun, as in (7).
(7) The student of the teacher who called you yesterday is sick.
However, the two nouns have the right features to function as antecedent of
the pronoun – they are both [þhuman] – and the interpretation is ambigu-
ous. Clearly, the preferred interpretation takes the closer antecedent (i.e.
teacher). In addition, if there is more available intervening material between
the head noun and the relative pronoun, the chances of acceptability of the
sentence are proportionally lowered.
The fact that there is no restriction on the function that the head noun
can have in the matrix clause is illustrated below. It can be a direct core
argument like in (6a) or it can even be an adverbial as in (8).
In this case the relativizer is not a pronoun but a wh-form with an adverbial
function in the RC; in other languages, it could be any pro-adverbial form.2
In English as well as many other languages there are no restrictions on the
function of the co-indexed proform in the relative clause, as suggested by (8).
However, some languages pose specific requirements. As noted by Keenan
(1976), Malagasy allows only the co-indexed element to be the subject of the
relative clause (9).
SENTENCE
CLAUSE
CORE
RP NUC RP
CORE PRED
NUCN PERIPHERY
CLAUSE
PrCS CORE
RP NUC RP
PRED
(10) INGR quit′ ([be′ (student1¬i, [INGR fail′ (who1, test)])], program)
----------
The regular syntax-to-semantics linking algorithm applies to relate
Figure 15.1 to (10) but, in addition, we need the following further restric-
tions (Van Valin 2012: 57).
(11) a. Retrieve from the lexicon an attributive LS (i.e. be′ (x, [pred′])) and
substitute the LS of the verb in the relative clause (i.e. [INGR fail′ (x, y)]) for
the second argument of the attributive LS (i.e. [pred′]).
b. Co-index the first argument in the attributive LS with either the unlinked
argument position in the relative clause LS, or, if there is a relative
pronoun, to the argument position linked to the relative pronoun.
c. Insert the attributive LS into the argument position in the matrix LS
occupied by the head noun, replacing the variable in the first argument
position in the attributive LS with the head noun.
There is still the question whether this structure complies with the
Completeness Constraint in cases like (12), where there is no overt relative
pronoun. The absence of an overt relative pronoun is possible in all cases
except in those where the pronoun is linked to the argument in the LS that
should be the PSA/subject.
In this case, the undergoer of see is not realized within the embedded
clause. This requires the reduction of a slot in the verb core, although the
structural conditions for this reduction are entirely different from those
that characterize control structures (Van Valin 2005: 255). Control
involves a non-subordination nexus at a core juncture, in which the
relevant core is a semantic argument of the matrix core. Quite differ-
ently, the relative clause in RC is a subordinated optional modifier of a
noun nucleus.
The answer to this problem is to resort to the formulation of the Com-
pleteness Constraint in Van Valin (2005: 233). Under this principle, the
argument of an LS that is not syntactically realized within the clause can
be satisfied anywhere within the sentence that contains that clause. In this
way Completeness is respected. Still, as is the case with control structures,
we need to determine the specific structural conditions that license this slot
reduction in RCs. The following is an attempt to capture the relevant
restrictions.
(13) The syntactic slots of the core of a finite clause may be reduced by 1 if the
argument to which the missing slot should have been linked is fulfilled
by a variable co-indexed with an antecedent noun that is the nucleus
modified by the phrase containing the empty slot.
In this way we identify the specific condition that allows the optional
reduction of an argument slot in some English EHRCs.
Internally headed relative clauses are characterized by the head noun
being part of the relative clause. In particular, the head noun is expressed
in a position within the relative clause, while it is omitted in the matrix
clause. This type of RC is found in Bambara, an SOV language.
(14) Bambara (Mande, Africa; Bird 1968, cited in Van Valin 2012)
[Ne ye so min ye] tye ye san.
1sg pst horse rel see man pst buy
‘The man bought the horse that I saw.’
The relevant question is how the hearer recovers the semantic function of
the co-indexed participant in the matrix LS. Upon hearing (14), a transitive
syntactic template is retrieved because of the verb buy. There is a relativizer
which indicates the presence of an embedded clause headed by the verb see,
which contains two arguments: ne (‘I’) and so (‘horse’). The word order of the
language (SOV) indicates that the first one is the Actor (and, hence, the PSA)
while the second is the Undergoer and, thus, the non-PSA macrorole argu-
ment. In turn, the matrix clause nucleus is buy, which requires two argu-
ments, although there is only one: tye (‘man’). The fact that the verb buy
selects human actors is the only cue that indicates, first, that man is the
Actor and, second, that there is a missing Undergoer argument, which has to
be found in the relative clause. There are two candidates, one is I and the
other is horse. World knowledge indicates that people are less likely to be
sold than horses and this might motivate the choice of horse in this case. In
other cases, the structure may be disambiguated on the basis of contextual
SENTENCE
CLAUSE
CORE
RP RP NUC
CLAUSE
CORE
RP RP NUC
(15a) [do′ (v, Ø)] CAUSE [BECOME have′ (v, w)] LEXICON
co-indexing (15b)
RC LS substitution (15c)
[do′ (tye, Ø)] CAUSE [BECOME have′ (tye, [be′ (xi, [see′ (ne, soi)])])]
Figure 15.2 Internally headed relative clauses (from Van Valin 2012)
clues. Taking into account the preceding considerations, the rules governing
linking from syntax to semantics in IHRCs are the following (Van Valin
2012: 60):
(15) a. Retrieve from the lexicon an attributive LS and substitute the LS of the
verb in the relative clause for the second argument.
b. Co-index the first argument in the attributive LS with the argument in the
relative clause LS identified as the head noun.
c. Insert the attributive LS into the open argument position in the matrix LS.
(16) The dog bit the boy that killed the spider.
The antecedent boy is an Undergoer in the matrix clause. Yet, the co-indexed
relative pronoun is an Actor in the embedded clause. Therefore, the higher
status of the antecedent bears on the LS that contains it in relation to the LS
that includes the proform.
The Bambara IHRC is quite different in this regard. The antecedent is
within a clause that is at the same syntactic level as the reduced slot. In
turn, the semantics contains the bounded variable in the LS of the depend-
ent verb, which is nested within the attributive structure that contains the
antecedent as an argument. This contrast – plus the fact that the relative
clause precedes the matrix clause – strongly suggests that the co-indexing
relation in Bambara IHRCs is driven by pragmatics and world knowledge, as
we argued above.5
As for non-restrictive relative clauses, these are clearly different from
restrictive ones, in that there is a pause separating the relative clause from
its antecedent. In addition, the attribution does not serve the purpose of
helping the hearer to identify the referent. Indeed, as can be seen in (17), the
antecedent can be a proper noun. Consequently, the information it provides
does not need to be presupposed. These properties are well taken care of by
the analysis in Figure 15.3, in which the relative clause is a peripheral
modifier of the RP (Van Valin 2005: 222).
RP Periphery
CLAUSE
N PrCs CORE
RP NUC RP
CORE
NUC
SENTENCE
CLAUSE
CORE
CLAUSE V PP
CORE
NUC RP
Cleft sentences are as complex as they are interesting and, similarly to RCs,
they involve a matrix and a subordinate clause. There are different subtypes
of cleft, but in this chapter the focus will be on ‘it-clefts’, as represented by
sentence (21). This example includes a cleft clause (who blamed the bus driver)
introduced by a wh-pronoun that obtains its reference from the antecedent
noun (teacher) in the matrix clause, also called ‘clefted noun’. This matrix
clause consists of a copula with dummy it.
The wh-form that introduces the cleft clause can fulfil any direct core
argument position. In fact, it can even express adjuncts, as can be seen
in (22).
incomplete content of the cleft clause. The single proposition has been
divided into an open proposition – a representation that comes with an
open slot or variable – expressed by the clefted clause and a matrix clause
that provides the value for that variable. The cleft clause is part of the
presupposition and, hence, outside the potential focus domain in a strong
sense: neither its internal constituents taken individually nor the entire
clause as a unit can be the focus. Example (21), repeated in (26a) for conveni-
ence, cannot be a felicitous answer to the question What happened?. The
potential focus domain only has scope over the matrix clause as confirmed
by focus sensitive-particles. The negative operator in (26b) ranges only over
the matrix clause and, consequently, on the co-indexation of the teacher as
the Actor of the embedded clause.
Further, a yes/no question can only ask for the clefted RP. Was it the teacher
who blamed the bus driver? can be answered by (26a) whereas Was it the bus
driver who was blamed by the teacher cannot. The appropriate context for an it-
cleft sentence is a contrastive one; it had been asserted that somebody else
other than the teacher (for example, a student) had accused the bus driver.
This contrastive focus typically conveys an exhaustive reading of the clefted
RP; namely, it was the teacher and nobody else who accused the bus driver
(Kiss 1998).
Languages like French have constraints preventing PSAs – subjects – to be
part of the actual focus domain in a sentence focus structure (SF). Since
French also has rigid SVO word order, preverbal elements cannot normally
be in focus. In consequence, French might use clefts as in (27) or presenta-
tional clefts as in (28) as pragmatically felicitous in a context that demands
SF (Lambrecht 1994: 226, 2000: 653; Van Valin 1999: 519).
combination occurs: nuclear, core or clausal (see Chapter 13). Since clefts
contain an obligatory co-indexation relation among members of different
clauses, the corresponding juncture is at the core layer. Regarding nexus,
we can rule out daughter subordination since the subordinate clause does
not fill an argument slot in the matrix clause and the co-indexation rela-
tion might involve adjuncts in the embedded clause. Furthermore, the
embedded verb inflexion agrees with the wh-form and the clefted element
in the matrix clause does not control agreement. This can be seen in
Spanish where the antecedent noun is a first-person pronoun while the
embedded verb agrees with the third person that corresponds to the
embedded pronoun (Pavey 2004: 29).
SENTENCE
CLAUSE
CORE PERIPHERY
RP NUC CLAUSE
AUX RP NUC RP
so that the assertion falls directly on the identity relation between the
copula and the cleft sentence, which is the focus of the assertion.8
Spanish does not have It-clefts but it does have a construction that is
structurally similar to the English one. It places the focus on the cleft
constituent in a clause with a copula followed by an embedded clause
introduced by a relative pronoun.
A difference with English is that in the Spanish cleft, the clefted noun might
be the subject of the matrix clause – as in (32) – since it controls the
agreement with the verb inflexion (i.e. the copula). Hence, the absence of
an ‘it’ pronoun cannot be attributed to the Spanish ‘pro-drop’ status. Despite
the variety of relative clauses existing in Spanish, here the relative pronoun
has to be of a certain kind – that is, lo que (‘the.m.sg that’), la que (‘the.f.sg
that’), los que (‘the.pl that’) and so on. As in English, the morphosyntactic
realization of the antecedent noun is determined by the meaning of the
embedded verb. In (33) the subject is impersonal and the antecedent is the
nominal element of a non-predicative PP that marks indirect objects.
There is a sense in which the central meanings of it-clefts and RCs are
equivalent in terms of truth conditions. Specifically, in the sentence The
student that failed the test quit the program, the truth-conditional content of the
RC can also be expressed by a simple sentence (The student failed the test) and
this exact content can be expressed by an it-cleft (It was the student who failed
the test). The semantic difference in truth-conditional terms is reduced to the
minimum; it-clefts add to the interpretation of the relevant noun the fur-
ther assumption that the student was the only member who performed the
action (exhaustive reading). The bulk of the semantic contrast lies in the way
the semantic representation is assembled and these different assemblages
are the encodings of different communicative functions. Speakers use RCs to
provide further information that will help hearers to identify the referent of
a head noun. It-clefts, instead, highlight the identity of one participant with
a particular role in one event.
RRG captures this divergence through an independently motivated dis-
tinction. It is assumed that the be′ predicate includes different kinds of
predications (Van Valin 2005: 48), even though they might all be expressed
via a copula verb in a particular language. Two of them are relevant here;
one is attributive predication and the other is specification (Declerck 1988: 47;
Pavey 2004: 29), represented in (36) and (37), respectively. The former states a
predication relation between a predicate and a referential expression. It
subsumes attributive adjectives and, thus, the relation between relative
clauses and their head nouns. In contrast, the specification relation co-
indexes two referential expressions; one might be a constant (i.e. Chris) and
the other is a definite description (the winner), a representation that allows us
to identify a unique individual in the world.
The predicative structure (i.e. be′) that corresponds to the semantic relation
between the head noun and the relative clause is inserted into one of the
matrix LS argument slots (i.e. the first argument of quit′).The participant
that fills in a semantic slot is literally shared with the predicative LS. This is
not a typical co-indexation relation but a unique slot that belongs to two LSs,
which is shown with a thick broken underline, as was pointed out above. In
addition, this position is co-indexed with the argument slot filled by the wh-
word in the relative clause LS.
(40) It was all the passengers who had committed the murder.
(41) All the students who attended will receive a bonus point.
In (40) the quantifier has scope over only the clefted noun, and the co-
indexed pronoun in the cleft clause inherits this interpretation. The full
RP denotes the set of ‘all passengers’ and all of them committed the murder.
In other words, the cleft clause does not affect the interpretation of the
noun phrase in the matrix clause. By contrast, in (41), the quantifier has
scope over the RP the students, which has already been restricted by the
relative clause. Therefore, the relevant set does not include all the (context-
ually possible) students but only the subset of them that attended. Indeed, it
is implicated that some students did not attend the class. The relative clause
restricts the set of the noun to a subset and it is only to this subset that the
universal quantifier applies. These properties can be derived naturally by
representing relative clauses within the RP whereas cleft sentences are
outside the RP, as shown in Figure 15.6.
With respect to information structure, neither the relative clause in RC
nor the cleft clause in it-clefts are part of the assertion, but are instead
presupposed. They both carry a proposition with a variable, namely, an open
proposition. However, these embedded clauses differ in subtle ways. The
relative clause helps the interlocutor to identify the reference of the head
RP, which is necessarily a non-identifiable participant. In contrast, the
In sum, the relative clauses in RC can be the focus as a whole because they
are nuclear modifiers of the RP, which might be within the potential focus
domain of the sentence. If the relative clause modifies a noun head that is
outside the potential focus domain – for example, an RP within an adverbial
clause in a clause juncture – the relative clause loses the possibility of being
part of the focus.
15.5 Conclusion
References
Bentley, Delia, Francesco Maria Ciconte and Silvio Cruschina. 2015. Existentials
and Locatives in Romance Dialects of Italy. Oxford: Oxford University Press.
Bird, Charles. 1968. Relative clauses in Bambara. Journal of West African
Languages 5: 35–47.
Chafe, Wallace. 1987. Cognitive constraints on information flow. In Robert
Tomlin (ed.), Coherence and Grounding in Discourse (Typological Studies in
Language 11), 21–52. Amsterdam: John Benjamins.
Cinque, Guglielmo. 2015. Three phenomena discriminating between ‘rais-
ing’ and ‘matching’ relative clauses. Semantics–Syntax Interface 2(1): 1–26.
Citko, Barbara. 2004. On headed, headless and light-headed relatives. Natural
Language and Linguistic Theory 22(1): 95–126.
Comrie, Bernard. 1998. Rethinking the typology of relative clauses. Language
Design 1: 59–86.
Davidse, Kristin. 2000. A constructional approach to clefts. Linguistics 38(6):
1101–1131.
Declerck, Renaat. 1988. Studies on Copular Sentences, Clefts and Pseudo-Clefts.
Berlin: De Gruyter Mouton.
Notes
(i) Subject > DO > IO > OBL > GEN(itive) > O(blique)COMP
(i) [see′ (ne, soi)] ^ [do′ (tye, Ø)] CAUSE [BECOME have′ (tye, xi)]
Abbreviations
16.1 Introduction
* I thank Delia Bentley and Robert D. Van Valin, Jr. for their valuable comments on earlier versions of the chapter.
(1) a. Jon believes [NPthe rumour [Sthat Mary lost her phone]].
b. *What does Jon believe [NPthe rumour [Sthat Mary lost __ ]? (complex NP)
c. Jon talked to [NPthe neighbour [Swho bought the house recently]].
d. *What did Jon talk to [NPthe neighbour [Swho bought __ recently]]? (complex NP)
e. Jon ate a sandwich after [SMary made an omelette].
f. *What did [SJon eat a sandwich after [SMay made __ ]]? (adjunct)
g. [SThat Jon ate the sandwich] was obvious.
h. *What was [S[Sthat Jon ate __ ] obvious]? (subject)
In this formulation, it has been assumed that these constraints are purely
structural with no reference to a semantic, pragmatic, or other basis. There-
fore, they have been taken as evidence of autonomous syntax and ultimately
of a theory of universal grammar.
While there has been much theory-internal debate with respect to the
status of the subjacency condition (see Yoshimura 1992 and references cited
therein), there have also been proposals to account for the constraints,
external to the syntactocentric theories, such as semantic and pragmatic
approaches (Erteschik-Shir 1973; Erteschik-Shir and Lappin 1979; Haig 1996;
Kuno 1987; Kuno and Takami 1993; Shimojo 2002; Van Valin 1996; Van Valin
and LaPolla 1997), cognitive and performance-based proposals (Deane 1991,
1992; Hofmeister 2007; Kluender 1990; Kluender and Kutas 1993), and pars-
ing and expectation-based claims (Chaves 2013). In short, the subjacency
accounts are problematic because the constraints are sensitive to a range of
factors beyond pure syntactic grounds. Furthermore, for a number of theor-
ies, the subjacency account is problematic since it assumes movement,
which is a theory-internal assumption, and therefore incompatible with
monostratal and non-transformational theories such as RRG.
In addition, there is difficulty with a movement-based approach in
accounting for constraints which do not involve movement without added
theory-internal stipulations. Lakhota is a language that does not favour the
subjacency account. Van Valin (1996: 36–37) points out that Lakhota blocks
wh-questions formed within a complex NP despite its wh-in-situ characteris-
tics. A sentence containing a word such as táku or tuwá and the question
SENTENCE
CLAUSE
CORE
RP NUC RP
V CORE PERIPHERY
PRED ADV
IU IU IU [IU IU IU IU]
IU
SPEECH ACT
Figure 16.1 Potential focus domain in clausal (daughter) subordination (Van Valin 2005: 214)
SENTENCE
CLAUSE
CORE PERIPHERY
RP NUC RP PP
V P CORE
PRED
IU IU IU IU
SPEECH ACT
Figure 16.2 Potential focus domain in ad-core subordination (Van Valin 2005: 216)
(5) Q: Did Pat see Kim after shei arrived at the party?
A: a. No, Sally.
b. No, before.
c. *No, shei left.2
The cross-linguistic variations outlined thus far raise a question about the
validity of the RRG principles in a language in which extraction is
allowed more freely, and therefore, this section discusses extraction
restrictions in Japanese. Studies on extraction constraints have been
centred on English-type languages, in which extraction constraints are
more or less structurally predictable. In relatively island-free languages,
restrictions are typically described on a semantic and pragmatic basis,
and Japanese is no exception (see the references cited for these languages
in Section 16.1).
As Van Valin and LaPolla (1997: 626) point out, in English, an adjunct wh-
question is ambiguous in a complex sentence in which the PFD extends over
the subordinate clause, as shown in (15a), because ‘when’ can be interpreted
as modifying either the matrix clause or the object complement. However,
ambiguity does not arise in a subordinate clause which is outside a PFD, as
in (15b).
(15) a. Wheni/j did Skinner say __i that Krycek would be at the missile silo __j?
b. Wheni/*j did Skully interview the witness __i who saw the alien spacecraft
in the silo __j?
On the one hand, the PFD excludes the pre-detached position (PrDP) and
the post-detached position (PoDP), which contain a sentence-initial topic and
a sentence-final (i.e. post-predicative) topic respectively. On the other hand,
the focus domain includes the pre-core slot (PrCS), which contains a
sentence-initial narrow-focus (i.e. a narrow-focus subject or a pre-posed
narrow-focus), and the post-core slot (PoCS), which contains a post-
predicative focus. Figure 16.3 shows the layered structure of the clause with
respect to the PFD.
Japanese uses post-nominal markings that correlate with the information
structuring of the sentence. The topic of a sentence, if any, is outside the PFD
SENTENCE
RP RP NUC
PRED
SPEECH ACT
Figure 16.3 Layered structure of the clause and potential focus domain
and it may be overtly marked with the topic marker wa. In contrast, if a
nominative argument is marked with the nominative marker, it is within
the PFD, whether it is narrow focus (PrCS) or part of a broad-focus (RP). The
direct evidence for a detached position being outside of the PFD comes
from the observation that a wh-question word cannot be topicalized, as
shown in (17).4
The functional principle given in (6) predicts that an extraction site is not
permitted within a topic phrase, which is outside the PFD, and this is indeed
the case, as shown in (19). The sentence with the topicalized complex RP is
not felicitous when no particular presupposition for the sentence is present
(for example, when the sentence is given out of the blue).
This is also the case with relativization and topicalization out of a topica-
lized complex RP, as shown in (20) and (21) respectively.
publishes as opposed to books which others publish, and if those in (20) and
(21) can be interpreted as contrastive with respect to books as opposed to
other things. In fact, the imposed contrastiveness associated with an extrac-
tion site further supports the principle in (6).
Erteschik-Shir (2007) claims that contrast is contextually constrained to
occur only if a contrast set is available, and because contrast is represented
by singling out a subset of a whole to separate it from the remaining
subset, the singled-out subset represents a subordinate focus, which is
embedded in the matrix topic. An example is given in (22). ‘John’ is part
of the presupposition for (22A); however, it is a newly singled-out member
of the previously given set (hence it is focal). In the formal notation, a
contrastive topic represents the subordinate focus structure, as shown on
the right-hand side. In other words, a (matrix) topic which contains a focus
simultaneously represents a contrast, and this corroborates the principle
in (6) because a topic must be contrastive (i.e. focal) in order to contain an
extraction site.
Truncation of this example results in [hon-ga eegakasareta] jon ‘Jon who a book
was made into a movie’. Possible interpretations of this truncation are Jon as
the author, editor, translator, etc. of the book (as translated as ‘Jon whose
book was made into a movie’), not a mere reader, and this contradicts the
intended reading of [[ __ *yonda] hon-ga eegakasareta] jon.
16.3.3.2 Postposing
Sentence elements are also displaced through the so-called postposing
construction. While Japanese is considered a rigid verb-final language, it
allows sentence elements to be in post-predicative positions, typically in
spoken Japanese (Shimojo 1995).8 An example is shown in (31b), in which
the nominative argument appears in the postverbal position, in contrast
with (31a), which represents the canonical ordering of arguments (the post-
posed elements are underlined in the examples).
As shown in Figure 16.3, there are two positions for post-predicative elements
in Japanese, PoCS and PoDP. While both are structurally marked positions in
the verb-final language, they represent distinct focus-structure properties. The
PoDP is a post-predicative topic position, the post-predicative counterpart of
PrDP, and for this reason, the preceding discussion of topicalization applies
here. For the PoDP-type (topic) postposing, the preceding part of the sentence
including the matrix clause must be about the displaced element. The
examples in (32) show the correspondence between acceptable relativization
(a), topicalization (b) and PoDP-type postposing (c).
On the other hand, because a postverbal element in PoCS does not repre-
sent a topic, the aboutness condition is not a requirement for the PoCS-type
(focus) postposing.9 For example, the postposing in (33b) is acceptable des-
pite the unacceptability of the corresponding relativization.
However, it has been observed that this type of postposing is sensitive to the
focus structure of a sentence. Simon (1989) gives the examples in (34) to
point out that transparency defined by genericness correlates with the
gradient acceptability. The postposing is acceptable with a matrix copula
(34a), and it is least acceptable with a specific action verb (34c).
However, in the context given in (36A), the relative clause is outside the
actual focus domain, and the matrix (bridging) clause represents the narrow
focus, which clearly impairs its postposability.10
16.3.3.3 Wh-questions
While wh-questions in complex sentences in Japanese are generally accept-
able, some restrictions have been observed. Consider the examples in (38),
taken from Haig (1979: 90–91, grammaticality judgement original).
Again, if the foci of a question are kept contiguous as shown in (49b, c),
these questions are considerably more acceptable, if not perfectly
acceptable.
(50) a. jon-wa [taroo-ga naze kinoo sono hon-o katta tte] itta no?
John-top Taro-nom why yesterday that book-acc buy.pst comp say.pst fp
‘Did John say that Taro bought the book yesterday why?’
b. *jon-wa [taroo-ga naze kinoo katta] hon-o karita no?
John-top Taro-nom why yesterday buy.pst book-acc borrow.pst fp
‘Did John borrow the book which Taro bought yesterday why?’
c. *[taroo-ga naze kinoo sono hon-o yonda atode] shiken-ga
Taro-nom why yesterday that book-acc read.pst after exam-nom
umaku itta no?
well go.pst fp
‘Did the exam go well after Taro read the book yesterday why?’
On the other hand, the other wh-adverbials such as dooyatte ‘how’ and
nominal wh-adverbials such as {donna/dooyuu} riyuu-de ‘for what reason’ are
not subject to the restriction, as shown in (51).13
The contrast in acceptability between (50) and (51) suggests that the
observed restriction is due to the property of the particular wh-word why.
Van Valin (2002: 169) points out the adsentential nature of why and states
that the answers to who, what and how involve some change in the original
sentence (which is also the case with the nominal wh-adverbial for what
reason in (51b)), but the answer to a why question is normally a because-clause,
which is adjoined to the unchanged original sentence. Why by itself can be a
felicitous question when it follows a preceding statement, and the answer to
the question does not change or elaborate the preceding statement, as
shown in (52).
As Van Valin (2005: 284) explains, the interclausal ‘reason’ relation, [LS1]
BECAUSE′ [LS2], contains two independent logical structures, which, how-
ever, are linked by a subordinating conjunction. Therefore, there is only one
illocutionary operator for the two logical structures. For this reason, only
one of the clauses can be the potential focus domain at a time.14 This is
exemplified by (52A2), in which ‘Taro was late’ is in the potential focus
domain and ‘John was upset’ represents the presupposition. In a why ques-
tion, the wh-word, which corresponds with [LS2] of the interclausal semantic
relation above, must be in the potential focus domain since it is the focus of
the question. This means that the clause representing the other logical
structure must be outside the potential focus domain, necessarily represent-
ing presupposition.15
Then, why do we observe the extraction restriction on questions such as
those in (50b, c)? There is a simple Gricean explanation for this. Since the
non-wh-part of a why question must be presupposed, only the presuppos-
ition that is relevant to the point of the question must be expressed,
because of the maxim of relevance. In (50b), for example, the presuppos-
ition ‘Taro bought the book yesterday’ is relevant to the why question since
it represents one of the logical structures in the ‘reason’ relation, but ‘John
borrowed the book’ is not relevant because it is not part of the semantic
relation. Inclusion of the irrelevant proposition in the question must draw
the focus for the Gricean reason, which shifts the potential focus domain
away from the wh-expression. This is essentially the same Gricean explan-
ation which applies to the unacceptable questions in (40) discussed earlier.
With the account above, the acceptable why question formed in an object
complement such as (50a) is expected because the embedded clause is a
direct daughter of the matrix clause node (see the principle in (6)). In other
words, what is expressed by the embedded clause is required by and hence
directly relevant to the matrix clause. To summarize, the fundamental
functional requirement that the wh-expression must function in the part
of an utterance which represents the assertion is a valid principle for naze
‘why’ questions in Japanese as well.
16.4 Conclusion
because it is assumed (Van Valin 1996, 1998, 2005; Van Valin and LaPolla
1997) that the constraints are ultimately derivable from Grice’s (1975)
cooperative principle and the maxim of quantity, which are the general
principles of rational human behaviour (hence, a challenge to the subja-
cency principle as evidence of autonomous syntax, and ultimately to a
theory of universal grammar). Van Valin (1998) uses the following from
Kempson (1975: 190) to illustrate how the Gricean explanation is rooted in
the claim that the focus domain is the domain for extraction (i.e. question-
ing and predication).16
The speaker believes the hearer knows (and knows that the speaker
knows) a certain body of propositions (i.e. that there is a pragmatic
universe of discourse) and in making a certain utterance . . . he believes
that the hearer, knowing the conventions of the language and hence the
conditions for the truth of the proposition in question, will recognize a
subset of those conditions as being part of that pragmatic universe of
discourse and hence neither assertible, deniable or queriable (without
violating the quantity maxim), and a second mutually exclusive subset
of the conditions as being outside the pragmatic universe of discourse.
This latter set, he will interpret as being asserted, denied, commanded
or queried.
References
Notes
17.1 Introduction
This chapter will briefly review the work on diachronic syntax and morpho-
syntax in Role and Reference Grammar (RRG) and show how the tools of
language description developed by this theory can be used to account for
several aspects of language change. Although RRG was designed primarily as
a synchronic syntactic theory, it has nevertheless been successfully applied
in description, and sometimes explanation, of a number of diachronic
syntactic processes (for a survey, see Wiemer 2008a). It is unsurprising that
most diachronic work in RRG was devoted to Indo-European languages, as
their history is generally better known than the histories of most non-Indo-
European languages, but there are exceptions to this claim. Among dia-
chronic studies dealing with non-Indo-European languages in the RRG
framework we can mention Ohori’s dissertation (1992) and article on the
development of Old Japanese clause linkage (1994) and Wilkins’ account of
the diachronic development of verbs expressing associated motion in
Mparntwe Arrernte, an Aboriginal language of Australia (1991). Ardis
Eschenberg’s dissertation on the article system of Omaha (2005) also con-
tains important insights into its diachronic development. Within the Indo-
European family, especially well represented are diachronic studies of con-
temporary languages such as German (Diedrichsen 2008), French (Kailuweit
2008), English (Nicolle 2008) and Croatian (Matasović 2012), but aspects of
the histories of dead languages such as Latin have also been analysed in the
RRG framework (Cennamo 2001).
Since RRG is not a theory that lays much emphasis on the intuition of the
ideal speaker/hearer as the source of evidence for the grammaticality of
sentences, and invented ‘laboratory’ examples do not play a great role in the
argumentation about the syntactic structures of a language, it is easy to
analyse dead languages with limited corpora within the RRG framework.
RRG was therefore successfully used in the analysis of some aspects of syntax
approaches, which often explicitly claim that every change not affecting
the UG is equally probable (Kroch 2001: 726). Thus, for example, in RRG
it would be natural to claim that demonstratives are more likely to turn
into definite articles than into aspect markers on verbs, both because of
their similar function (establishing reference and marking definite ref-
erence) and because they have similar positions in the layered structure
of the reference phrase (RP). Such a claim amounts to a statistically
verifiable prediction which can be checked by looking at the history of
various languages in which the diachronic path from demonstratives to
definite articles can be observed. For example, in the Romance lan-
guages the articles developed from Latin demonstratives ille and iste, in
Ancient Greek the Homeric demonstrative ho became an article by the
Classical period, and similar developments can be observed in the his-
tory of a number of Balkan languages (e.g. Bulgarian and Albanian) and
Armenian (for other examples, see Matasović 2002: 59). The development
from demonstratives to aspect markers on verbs is, to the best of our
knowledge, unattested.1 A theory that cannot or does not make such
verifiable predictions is clearly weaker with respect to its predictive
power than RRG.
iii. Syntactic theories that presuppose the autonomy of syntax would tend
to predict that syntactic change is generally independent of changes in
other domains of grammar. RRG, as a theory that explicitly denies that
syntax is autonomous (in the generativist sense), sees syntactic change
in close interdependence with changes in phonology, morphology,
semantics and pragmatics. For example, in a number of languages it
can be observed that the emphatic pronoun, used for emphasizing (or
contrasting) the reference of a personal pronoun, becomes the reflexive
marker, used to establish obligatory coreference in the clause or dis-
course. This change, which occurred in the history of the English
reflexive (him-self etc.) as well as in Irish (féin), Ancient Greek (he-autón)
and other languages, is motivated by the semantic and pragmatic prox-
imity between emphasizing that a referent controls the action (rather
than someone else) and asserting that a referent acts upon himself/
herself (rather than upon someone else). Since the emphasizing pro-
noun, by virtue of its meaning and pragmatic use, has to have an
antecedent, it can also be used to establish coreference between itself
and its antecedent. Thus, in many languages the emphasizing and
reflexive markers are homophonous, as in the English sentences John
saw himself in the mirror (reflexive himself) and John closed the mirror himself
(emphasizing). The reverse process (from emphasizing to reflexive) is
unattested, as far as we know, since the use of a reflexive marker to
emphasize that a referent controls the action would not be pragmatic-
ally or semantically motivated.
iv. Generativist theories of syntactic change tend to assume that change
essentially takes place during language learning and is a consequence of
expressing the aspect of the verb. In a language that has only suffixes, this
will necessarily mean that the suffix for aspect is closer to the root than the
suffix expressing illocutionary force, for example the question marker. Now
a consequence of the universality of the NSP is that no syntactic change is
possible that would violate it, which means that, for example, if a language
already has suffixes expressing illocutionary force – or some other clause- or
sentence-level operators – no new suffixes expressing operators with a
narrower scope, such as aspect, can develop. Thus, in Indo-European lan-
guages that had developed tense markers expressed by suffixes, new aspect-
ual markers could not be grammaticalized as suffixes, but rather developed
as prefixes, for example in Old Irish, where the simple preterite car-ais (love-
3sg.pret.) ‘he loved’ is opposed to the new perfect ro-car-Ø (prf.-love-3sg.). In
other languages, the development was more complex, for example in Slavic,
where full aspectual oppositions developed only after the new prefixed
forms (e.g. Croatian po-gledati ‘take a look’) came into opposition with non-
prefixed (gledati ‘watch’) and both prefixed and suffixed forms (po-gleda-avati
‘watch repeatedly’), but in no Indo-European language can we observe a
development of new suffixes or postverbal particles expressing aspectual
meanings, as such morphemes would necessarily follow inflexional mor-
phemes expressing tense, thus violating the NSP (for details see Matasović
2002: 63–64 and Wiemer 2008a for some corrections).
The NSP plays a role in another generalization that has been suggested
concerning the frequency of grammaticalization patterns. It has been
argued that operators with narrower scope tend to grammaticalize as oper-
ators with broader scope, but not vice versa (Foley and Van Valin 1984:
216–217; Matasović 2008). This means that, for example, core operators such
as morphemes expressing deontic modality will develop into clausal oper-
ators such as those expressing status (epistemic modality), but that the
reverse pattern would be rare or unattested. An example of such a pattern
is the change of the English modal must from deontic (The letter must arrive
next week) to epistemic (The letter must be in the mail). Similarly, nuclear
operators such as those expressing aspect regularly grammaticalize as oper-
ators with scope over the clause or sentence, such as markers of evidenti-
ality. An example of this pattern is the development of the participle of the
auxiliary ima ‘have’ in Macedonian, which was originally just the marker of
perfective aspect, but it has developed into an evidentiality marker express-
ing the ‘hearsay’ meaning, for example in the sentence Ti si imal kupeno kniga
(you be.3sg.pret have.ptcp bought.ptcp book) ‘You bought a book (someone
told me)’. This empirical generalization, which certainly requires further
elaboration and empirical corroboration, is not directly implied by any
principles of RRG, but the NSP, which is one of the theory’s fundamental
principles, allowed its precise formulation. In theories which do not posit
the NSP or its equivalent, it would not be possible either to formulate this
generalization or to check it empirically in a systematic manner. However,
motivated topic and/or focus markers gradually lost their pragmatic func-
tions and became markers of PSAs. It remains to be seen whether this
general diachronic typology of PSA development can be extended to other
languages and language families.
There may well be other directional and universal patterns of diachronic
development that are precisely formulable in the RRG framework, but the
current level of our knowledge is insufficient to make strong hypotheses. For
example, in a number of ancient or archaic Indo-European languages the
reflexive pronouns derived from the Proto-Indo-European root *s(w)e- can be
used in the embedded clause to refer back to its antecedent in the main
clause. This is the case with Latin se in (2).
In terms of RRG’s layered structure of the clause we can say that the
domain of reflexivization in such languages is the clause, since the reflexive
marker is within the same (superordinate) clause, although it is not the
argument of the same verb as its antecedent (i.e. it is not in the same core).
On the other hand, in most modern Indo-European languages (including the
Romance languages, which developed from Latin), the antecedent of reflex-
ivization and the reflexive marker (derived from *s(w)e-) must be in the same
core,3 which means that the domain of reflexivization is the core, not the
clause. For example, the reflex of Latin se in French, the clitic se, cannot be
used in the subordinate clause as coreferent with the subject of the main
clause.4 It appears, then, that the archaic use of the reflexive marker found
in Latin was changed in a way that led to the narrowing of the domain of
reflexivization in many modern Indo-European languages (with the excep-
tion of Icelandic, which allows for ‘long-distance reflexivization’ but seems
to be archaic in this regard, cf. Van Valin and LaPolla 1997: 613 and Wack-
ernagel 2009: 515 for ‘long-distance reflexivization’ in Old Icelandic). It
would be interesting to know if there are attested instances of diachronic
developments of reflexive markers by which the domain of reflexivization in
a language was broadened, rather than narrowed. This is clearly a worth-
while topic for future research.
Since any aspect of the syntax and the levels of representation that are
mapped to it in the linking are subject to change, representing diachronic
In RRG terms, we would say that at this stage the following changes
occurred: (1) The inventory of syntactic templates was enriched with a
construction in which the topic was obligatorily copied in postverbal pos-
ition. The full pronouns were preserved in the pre-detached position in such
constructions (the default position for topics). (2) The focus structure was
affected, as the topical element now had to be postverbal. (3) By a subse-
quent change, the focus structure was affected again, as the clitics lost
topicality and became pure agreement markers expressing the lowest
macrorole argument of the verb (the absolutive argument) and the highest
argument of the verb (the ergative argument). This change affected the
syntactic templates of the language, as the clitics now became integrated
into the core.
In RRG terms, this means that the linking algorithm (case assignment for
clitics) was affected at this stage, in that the rule that applies in them
states that the highest macrorole (clitic) argument received the
‘nominative’ case, just as in Latin or any other nominative-accusative
language, and unlike in Tabassaran’s closest relatives. This is only a step
from a further change in which the clitics became fully integrated into
verbal morphology as person endings agreeing with the ‘subject’ (the
highest-ranking macrorole argument) of the verb. In this system – which
is perhaps how contemporary Northern Tabassaran should be analysed –
the language presumably acquired the head-marking feature by which the
verbal suffixes, rather than free pronouns, should be interpreted as argu-
ments, and this again affected the inventory of syntactic templates in the
language. Similar developments can be observed in other languages, in
which pronominal clitics, originally used in pragmatically marked con-
structions to mark direct and indirect objects, became agreement markers
on verbs. This is what happened in several Balkan languages with ‘clitic
doubling’, including Macedonian and Bulgarian, and a similar process can
be observed in a number of Spanish dialects (especially in Argentinian
Spanish, see Belloro 2004). Although in those languages the change from
clitics in pragmatically marked constructions to agreement markers on
verbs did not lead to alignment change, the initial steps of the process
were very similar to what happened in Tabassaran, as the syntactic tem-
plates and focus structure were similarly affected. Thus, the RRG approach
allows us to clearly see the similarities and differences in historical pro-
cesses occurring in very different languages.
It is a reasonable prediction that complex changes affecting several
aspects of syntax will be cross-linguistically much rarer than simple changes
affecting only a single syntactic structure or representation. This would
mean that, for example, clause-alignment changes will be much rarer,
cross-linguistically, than simple word-order changes, which affect only the
syntactic templates of a language and not the other aspects of its syntax.
However, other factors having little to do with pure syntax have to be taken
into account, for example the discovery of areal linguistics that both word
order and clause alignment are areally stable features of language, which
are therefore unlikely to change except in some historically extraordinary
circumstances (for example mass migrations of speakers of a language to a
different language area). Thus, the Indian Subcontinent has apparently been
characterized as a language area where the SOV languages have predomin-
ated for at least two thousand years and possibly much longer (Vedic
Sanskrit, the ancestor of all Indo-Aryan languages, was SOV, and Proto-
Dravidian was also probably SOV). However, Romani, which is spoken by
people who emigrated from Northern India roughly a thousand years ago, is
SVO, like Armenian, Greek and most other languages it was affected by on
the path from India to Europe.
Synchronic generalizations about universal, or quasi-universal aspects of
syntactic structure always have their bearing upon hypotheses about pos-
sible and probable patterns of change. Thus, it has been noted that the
universal aspects of clause structure (the nucleus, core, periphery and
clause) are all semantically motivated, while the non-universal aspects (the
detached phrases, the extra-core slots) are pragmatically motivated (in the
sense that they are associated with constructions that have strong pragmatic
restrictions on their occurrence, Van Valin 2005: 8). It should be noted that
these are empirical findings, not something postulated by the theory, but
they have led to the hypothesis that semantically motivated features of
language are diachronically stable, while pragmatically motivated features
are rather unstable and more likely to change in the history of a language.
Although this hypothesis appears to be motivated by the theoretical archi-
tecture of RRG, it has not been proved or subjected to systematic research
beyond a rather impressionistic overview of the literature on grammaticali-
zation (Matasović 2002).
A theory of diachronic syntax must also be able to capture different types
of syntactic change. Instances of syntactic change can generally be classified
as one of three types: reanalysis, extension, and borrowing (Harris and
Campbell 1995). Reanalysis is a process that involves potentially ambiguous
syntactic structures which receive new structural representation and are
parsed differently by language speakers before and after the change in
question. Extension is the process of generalization of rules or constructions
from one domain to others, and borrowing involves the adoption of con-
structions from one language into another. All three of these types of change
can be represented in RRG, and their representation in the RRG framework
allows the researcher to make a more fine-grained analysis than is usual in
general accounts of historical syntax.
A case of syntactic reanalysis followed by extension is presented in the
RRG framework by Diedrichsen (2008). She shows how in colloquial German
the nuclear juncture involving the verb bekommen ‘get’ and a number of
ditransitive verbs of transfer was reanalysed as a passive complex verb form.
An example of such a passive construction is (7):
Diedrichsen’s analysis allows us to see not only how reanalysis and exten-
sion work on the syntactic templates of the affected constructions, but also
how such syntactic changes slowly spread through the vocabulary of a
language, in a way that is affected by the lexical meanings of
individual items.
Borrowing of syntactic patterns can also be represented within the RRG
framework. Its existence and, indeed, widespread occurrence, do not pre-
sent a theoretical problem, since RRG does not make any theoretical
assumptions about when and how syntactic change takes place. As far as
the theory is concerned, syntactic change can occur both during language
acquisition, as a consequence of imperfect learning, and in adulthood,
either as a result of interference in the minds of bilingual speakers (syntactic
borrowing), or for some other reason. Here, again, the theory allows us to
represent different aspects of syntactic borrowing: firstly, borrowing can
involve the copying of specific syntactic templates as abstract patterns, for
example with respect to word order. This was the case when the SOV word
order was borrowed in Ethiopian Semitic languages such as Amharic from
neighbouring Cushitic languages such as Somali, or when the inversion rule
in the formation of yes/no questions was borrowed into French from Ger-
manic (Frankish), cf. French Vient-il? (comes-he) ‘Is he coming?’ with inver-
sion, which is also found in Dutch and German, but not in most
Romance languages.
Secondly, syntactic borrowing can involve the borrowing of individual
lexical items that has consequences for syntax, for example when a verb
such as German schmecken ‘to taste’ was borrowed into dialectal Croatian
(šmekati). Like German schmecken, which takes the dative experiencer and
the nominative source (e.g. EsNOM schmeckt mirDAT gut ‘It tastes good to me’),
Croatian dialectal šmekati has a quirky case frame (ToNOM miDAT dobro šmeka
‘It tastes good to me’). It remains to be seen whether such items can be
borrowed if the recipient language does not already have verbs with such
logical structures and case frames (Croatian has the inherited verb sviđati se
‘to like’, which has essentially the same syntax as the borrowed šmekati).
This also applies to instances where concrete forms are not borrowed but
the syntactic behaviour of individual lexical items is influenced by the
syntax of their translational equivalents in another language (syntactic
calques). For example, it is tempting to see the case frame of Croatian
darivati ‘present (imperfective)’ and darovati ‘present (perfective)’ as a syn-
tactic calque of Latin donare ‘present’, since both verbs show variable
undergoer selection, allowing both the theme argument and the recipient
to occur in the accusative (which is the default case of undergoers). This
means that two different thematic roles (with different positions on the
Actor-Undergoer Hierarchy) can be undergoers (Van Valin 2005: 111). The
Croatian verb darivati can take either accusative of the recipient and instru-
mental of the theme (darivati djecuACC kolačimaINS ‘to present children with
cakes’), or the dative of the recipient and the accusative of the theme
(darivati djeciDAT kolačeACC ‘to present cakes to children’), similarly to its Latin
equivalent which takes either dative and accusative (donare alicuiDAT
aliquidACC ‘to present something to someone’) or accusative and ablative
(donare aliquemACC aliquareABL ‘to present someone with something’; Latin
ablative is the usual equivalent of the Slavic instrumental case). However,
Croatian darivati has etymologically cognate verbs in other Slavic
languages, including Russian (darivat’), which also have two pairs of case
frames parallel to Croatian, which makes the hypothesis that those case
frames are inherited from Proto-Slavic more likely than the hypothesis that
they were independently borrowed from Latin. Inheritance seems even
more likely in light of the fact that verbs with such case frames are
otherwise almost unattested in Slavic: in Croatian, the only other example
is nuditi ‘offer’, which can also be construed with both the accusative and
the dative (nuditi djeciDAT kolačeACC ‘to offer the cakes to children’) and the
accusative and the instrumental (nuditi djecuACC kolačimaINS ‘to offer children
with cakes’). In the case of nuditi, a plausible explanation of its exceptional
case frame and variable undergoer selection properties would be that it is
the result of extension from darivati, which is inherited. The Latin verb
meaning ‘to offer’ (offerre) could not have been the model for nuditi, as it can
be construed only with the accusative and the dative (offerre aliquidACC
alicuiDAT ‘to offer something to someone’).
It seems fair to say that most lexical borrowing involves syntactic adapta-
tion to pre-existing patterns: for example, one does not expect to find cases
where verbs with ergative case frames are borrowed into a language with
nominative-accusative morphosyntax, or verbs showing person agreement
borrowed into a language with no person agreement. However, there are
documented cases of borrowing of lexical items that either introduce new
syntactic patterns into a language or contradict pre-existing ones. For
example, Tagalog, a genderless language, has borrowed adjectives from
Spanish, and those borrowed adjectives must agree in gender with their
head nouns in the RP (Schachter and Otanes 1972: 166–168; Matasović 2014);
the same rule holds for adjectives borrowed from Sanskrit into Marathi.
Conversely, Croatian, in which gender agreement in the RP is obligatory, has
borrowed a number of adjectives from English that show no gender agree-
ment (e.g. super ‘great, super’ and cool).
Although borrowing is easily accounted for and represented in RRG, it is
unclear whether there are constraints on grammatical borrowing in this
theory, and it remains to be seen if the theory offers any clues to such
potential constraints. For example, the borrowability of elements repre-
sented on the operator projection would need to be investigated. Research
into language contact has long shown that grammatical morphemes –
including inflexional elements – can be borrowed, though usually on a
smaller scale than lexical morphemes, but it is unclear whether there is a
difference of borrowability of different classes of grammatical morphemes.
Since RRG explicitly distinguishes two classes of grammatical elements with
respect to their syntactic representation, mapping some grammatical mor-
phemes in the constituent projection (e.g. causative markers, or person
markers of verbs), and others on the operator projection (e.g. aspect, tense
and illocutionary force markers), it would be interesting to investigate
whether these two classes differ with respect to borrowability. Again, it
seems that borrowing syntactic templates – which may be influenced by
pragmatic factors – is much more common than borrowing of elements
representable on the operator projection, which is strongly semantically
motivated. However, besides intuitions based on careful reading of litera-
ture on language contact, there is at present little systematic research to
corroborate such a claim.
Finally, in this chapter we have discussed the implications of RRG for a
theory of diachronic syntax, but RRG is actually more a general theory of
linguistic structure than a simple syntactic theory, since the interface of
syntax, semantics and pragmatics plays a central role in its architecture and
explanatory goals. Therefore, everything that has been claimed here about
historical syntax also has consequences for historical semantics and prag-
matics, and our discussion has shown a number of such consequences.
Moreover, the boundaries between morphology and syntax are set some-
what differently in RRG in comparison to other theories, so diachronic RRG
is also in part a theory of historical morphology. It has been established that
most bound grammatical morphemes were in origin independent words
with syntactic functions (‘Today’s morphology is yesterday’s syntax’, Givón
1971: 413), and this means that the change of their grammatical status will
often affect how they are represented in RRG. Thus, many Classical Greek
directional verbal prefixes (represented as nuclear operators in the operator
projection) were still independent adverbs in Homeric Greek, where they
would be represented as elements of the periphery in the constituent pro-
jection. For example, in Homeric Greek epı̀ ‘over, above’ is still separated
from the verb teínō ‘extend’, as in (9):
However, by the Classical period (5th century BCE) we find it only as the
prefix in epiteínō ‘extend over, cover’ (Lindemann and Färber 2003: 47).
On the other hand, not all diachronic developments that involve a free
morpheme becoming bound need affect the way those morphemes are repre-
sented in RRG. For example, the clitic pronouns of Literary French, such as the
je ‘I’ (from Latin ego), have largely become bound person-marking prefixes in
colloquial French, but in both stages of its development je would probably be
represented as an argument in the constituent projection. Only if the gram-
maticalization process is taken a step further, and je becomes a pure person-
agreement marking on the verb, while the former topicalized stressed pro-
noun (moi) changes its status and becomes the pragmatically unmarked
subject pronoun (like English I or German ich), would the way the morpheme
je is represented change, as moi would be linked to the argument position in
the constituent projection (rather than being represented as the topicalized
element in the pre-detached position, as in (10)).
17.4 Conclusion
It has been argued that Role and Reference Grammar provides useful tools
in the investigation of diachronic processes affecting the syntactic structure
of languages of various types. The fact that RRG is designed to be descrip-
tively adequate in representing typologically very diverse living languages,
and that its adequacy can be tested empirically on languages with unlimited
corpora and living native speakers should mean that this theory is also an
adequate tool for description of dead languages with limited corpora and no
native speakers. Describing two synchronic stages of a single language
within a uniform theoretical framework is a prerequisite for the objective
analysis of the changes that occurred from one stage to the other, and it has
been argued here that RRG is up to that task. We have also claimed that the
theory makes interesting predictions about the probability of certain types
of diachronic changes that can be tested empirically. Moreover, the archi-
tecture of the theory enables the investigator to formulate specific hypoth-
eses about the directionality of changes on particular levels of syntactic
structure, and we believe that fruitful applications of RRG lie in proposing
and testing such hypotheses, for example with respect to the diachronic
widening or narrowing of the scope of operators, the expanding or shrink-
ing domains of certain syntactic processes such as reflexive binding or
agreement, and the interaction of the focus structure domains with the
domains of purely syntactic processes. As the review of the published dia-
chronic applications of RRG testifies, this theory is also well designed to cope
with the description and explanation of cross-linguistically common
grammaticalization patterns.
Unfortunately, there has not been a lot of work in diachronic syntax
within the framework of RRG, but it is to be hoped that historical linguists
will realize the potential of this theory in the future.
References
Belloro, Valeria. 2004. A Role and Reference Grammar Account of Third-Person Clitic
Clusters in Spanish. Unpublished MA thesis, University at Buffalo (SUNY)
[available on the RRG website, https://rrg.caset.buffalo.edu/].
Cennamo, Michela. 2001. On the reorganization of voice distinctions and
grammatical relations in late Latin. In Claude Moussy (ed.), De Lingva
Latina Novae Questiones (Actes du Xème Colloque International de Linguistique
Latine, Paris-Sèvres, 19–23 avril 1999), 51–65. Louvain: Peeters.
Diedrichsen, Elke. 2008. The grammaticalization of the bekommen-passive in
a RRG-perspective. In Kailuweit et al. (eds.), 87–145.
Dryer, Matthew S. 1992. The Greenbergian word order correlations. Language
68: 81–138.
Eschenberg, Ardis. 2005. The Article System of Umonhon (Omaha). PhD disserta-
tion, University at Buffalo (SUNY) [available on the RRG website, https://
rrg.caset.buffalo.edu/].
Foley, William A. and Robert D. Van Valin, Jr. 1984. Functional Syntax and
Universal Grammar. Cambridge: Cambridge University Press.
Gildersleeve, Basil L. and Gonzalez Lodge. 1992. Latin Grammar. Edinburgh:
Nelson.
Givón, Talmy. 1971. Historical syntax and synchronic morphology: An
archaeologist’s field trip. Chicago Linguistic Society 7(1): 394–415.
Gonzalez Orta, Maria M. 2002. Lexical templates and syntactic variation: The
syntax–semantics interface of the Old English speech verb secgan. In Ricardo
Mairal Usón and Miguel J. Pérez Quintero (eds.), New Perspectives on Argument
Structure in Functional Grammar, 291–302. Amsterdam: John Benjamins.
Harris, Alice C. and Lyle Campbell. 1995. Historical Syntax in Cross-Linguistic
Perspective. Cambridge: Cambridge University Press.
Kailuweit, Rolf. 2008. Some remarks on RRG and grammaticalization:
French verbal periphrases. In Kailuweit et al. (eds.), 69–86.
Kailuweit, Rolf et al. (eds.). 2008. New Applications of Role and Reference Gram-
mar. Newcastle upon Tyne: Cambridge Scholars Publishing.
Kroch, Anthony S. 2001. Syntactic change. In Mark Baltin and Chris Collins
(eds.), Handbook of Contemporary Syntactic Theory, 699–729. Oxford: Oxford
University Press.
Notes
1 This does not mean that such a development is entirely impossible, only
that it is very improbable. Rare patterns of grammaticalization occasion-
ally lead to unexpected homonymy of grammatical morphemes. In
Omaha (a Siouan language) there is a set of articles which are also used
as auxiliary verbs and evidential markers, and Eschenberg (2005: 176–206)
convincingly shows how original copulas expressing orientation of their
subject were reanalysed as classificatory articles. For example, Webaxu khe.
Zi. (pencil be.horizontal yellow) ‘The pencil is positioned horizontally. It is
yellow’ was reanalysed as Webaxu-khe zi (pencil-definite yellow) ‘The pencil
is yellow’ (Eschenberg 2005: 183).
2 We use the Leipzig abbreviations (https://www.eva.mpg.de/lingua/resources/
glossing-rules.php) with the following addition: pret ¼ preterite.
3 Or the simple clause including the core and the pre-core slot (Van Valin
2005: 167).
4 I.e. one cannot say something like *Jeani pense que Pierre sei frappe (lit. Jean
thinks that Pierre himself is.hitting) ‘Jeani thinks that Pierre is hitting
himselfi’. French works exactly like English in this regard, but
unlike Latin.
Richard M. Weist
18.1 Introduction
The three research projects reviewed in this section share the following
properties: (1) they base their arguments on child language data, (2) they
propose that children understand predicate-argument structure that incorp-
orates semantic relations, (3) they do not find evidence to support a
Chomsky-type theory of the acquisition process (see, for example, Chomsky
1965 or Pinker 1984), and (4) they share with Van Valin (1993: 2), ‘the
conviction that grammatical structure can only be understood with refer-
ence to its semantic and communicative function’. However, none of the
investigators credit the toddler with the capacity to process inflectional
morphology (cf. the research on tense, aspect, and modality described in
Section 18.4).
Limited-Scope Formulae. In his seminal study of eleven children (20 to 26
months old) encompassing five different languages, Braine (1976: 4) pre-
sented the argument ‘that the first productive structures are formulae of
limited scope for realizing specific kinds of meanings’. According to Braine’s
observations, a limited-scope formula was ‘a rule that maps elements of a
semantic representation into positions in the surface structure’ (p. 69). Five
of the ten patterns summarized by Braine (pp. 56 and 57) are as follows, with
a semantic value linked to the position pattern: (1) identification, ‘it/that’ þ X;
(2) disappearance, ‘allgone’ þ X; (3) negation, ‘no’ þ X; (4) actor–action rela-
tions, Kendall swim, and (5) requests, ‘want/have-it’ þ X. Braine argued that
‘ALL early learning can be interpreted as a learning of position patterns’
(p. 66) and that the child language data do not support theories that
attribute grammatical structure such as S ! NP þ VP and VP ! V þ NP to
the language learner.1 In Section 18.6, I will return to an attempt to explain
the emergence of wh-question formation in English with a Braine-type
concept of ‘lexically-specific formulae’ (Rowland and Pine 2000: 157).
‘Case’ (or Semantic Function) Grammar. Bowerman (1973) investigated
the early stages of word combinations in two children learning Finnish,
Seppo and Rina, and one child learning English, Kendall. Citing Chomsky
(1965), Bowerman summarized that ‘the constituent structure assigned to
children’s utterances by transformational generative grammar is largely
gratuitous’ (p. 222) and specifically, ‘sentence-subjects tend initially to be
restricted largely to the semantic function ‘agent’ with a hand full of
exceptions for some children’ (p. 189). As an alternative, Bowerman
proposed a ‘case’ grammar based on the semantic functions agent (i.e. the
instigator of action) and object (i.e. the entity affected by action or state
identified) (see Fillmore 1968). In fact, Kendall and Seppo produced excep-
tions to the ‘agent-as-subject’ regularity, for example, Kendall’s pillow fell and
thread break and Seppo’s kissa putto ‘cat falls’, hirri pelä(sty) ‘mouse is-afraid’,
and torni kaatuu ‘tower falls-down’. Furthermore, the agent/subject is not
always animate, for example, helikopteri lentää ‘helicopter flies’. These ‘excep-
tions’ may preview the children’s understanding of more abstract grammat-
ical functions. The data indicate that children know more than ‘limited-
scope’ mapping rules (as Braine claimed), but the children were not yet
credited with the capacity to process morphosyntactic information (cf.
Toivainen 1980 for Finnish).
Usage-Based Grammar and Verb-Island Constructions. Tomasello (1992)
focused on the manner in which verbs emerged in the corpus of his daugh-
ter Travis from a cognitive-cultural perspective. According to Tomasello
(1992: 23), ‘young children’s early verbs and relational terms are individual
islands of organization in an otherwise unorganized grammatical system’,
and further, ‘In the early stages, the child learns about arguments and
syntactic marking on a verb-by-verb basis, and ordering patterns and mor-
phological markers learned for one verb do not immediately generalize to
other verbs.’ In contrast to the supposition that language emerges in the
child guided by innate universal grammar, Tomasello (2003: 21–31) pro-
posed that language is acquired by utilizing the principles of ‘intention-
reading’ and ‘pattern-finding’. According to the theory, children learn a set
of verb–argument constructions having a set of ‘slots’ to be filled with
specific semantic relations, for example, the schema ‘X made (this) Y’ was
realized as, Maria made this duck and Linda made ice. Before the age of 3,
Tomasello (2000: 71) found no motivation for productive verb categories or
inflectional morphemes. Hence, the verb in the construction [X singing]
would not be classified as an ‘activity’ verb, and the verb in [X fall-down]
would not be an ‘achievement’ verb. Further, neither the -ed in spilled Weezer
milk (Travis, 1;9) nor the -ing in Weezer drinking the eggs (Travis 1;7) were
judged to be productive morphemes coding tense and aspect because there
was very little evidence for contrast at this phase of development (cf. a rare
contrast: came off Grover (1;9) versus smoke coming out the chimney (1;11) (see
Tomasello 1992, Appendix). Contrary to Tomasello’s predictions, cross-
linguistic findings reveal the early influence of semantic relations on verb
categories and the rapid emergence of contrasts in the finite morphology
(see Tables 18.4 and 18.11).
Theories of Early Combinations and RRG. In his paper relating functional
linguistic theory to language acquisition, Van Valin (1991) presented an
argument for the relevance of RRG to child language research. Van Valin
began with the distinction between the ‘adaptationist’ and the ‘construc-
tionist’ views of language acquisition. Given the adaptationist view, the
child is innately programmed with a set of universal principles, and the
child sets parameters to conform to the target language (see, for example,
Pinker 1984, Radford 1990, or Sano and Hyams 1994). Alternatively,
according to the constructionist view, children learn language by utilizing
cognitive structures such as linguistic information processing principles
(see, for example, Slobin’s (1985) theory of ‘Operating Principles’). In their
analysis of single and early multiple word utterances, child language
researchers have observed the critical role that semantic and pragmatic
principles play in their data. The most fundamental principles of acquisition
that are derived from the child language data are also theoretical primitives
in the RRG theory. Thus, the RRG theoretical framework has the potential to
provide important insights into the question of how language is acquired.
This chapter will explore RRG’s potential in this regard.
Table 18.1 Layered structure of the clause (LSC) and operator scope in RRG
INGR (or instantaneous change) or BECOME (or change over time) in their
logical structure have the telic property. The course of language acquisition
is shaped by the interaction of the semantic structure of predicates with the
layered clause structure and its operators. Cross-linguistic research has
revealed a pervasive interaction of lexical and grammatical aspects. The
next few sections of this chapter will probe the child language data that
are relevant to this interaction revealing links to the RRG theoretical
framework.
Syntactic Realization of Logical Structure. Thematic relations are the
semantic roles/functions of the arguments in logical structure (see Van Valin
2005: 58, Fig. 2.3). The linking system to syntax involves two semantic
macroroles, Actor and Undergoer. These macroroles are generalizations
across argument types: Actor across agent-type and Undergoer across
patient-type thematic relations. The agent argument of DO is the least
marked choice for Actor, followed by the first argument of do′ (x), while it
is the most marked choice for Undergoer, while the patient/entity argument
of state pred′ (x) is the least marked choice for Undergoer, followed by the
second argument of pred′ (x, y), and the most marked choice for Actor. The
Actor-Undergoer Hierarchy specifies the mapping of thematically linked
arguments to macroroles, and, depending on language typology, mapping
continues to the privileged syntactic argument (PSA), which is the RRG
equivalent to syntactic subject (Van Valin, 2005: 127, Fig. 5.1). In a
nominative-accusative language like English, Polish or Japanese, the PSA
will be in the nominative case.
large variety of languages. From the RRG perspective, the acquisition data
provide insights into the relationship between the predicate that shapes
clause structure and the operators that modify levels of clause structure.
Predicate Structure and Operators in Turkish. I begin with a study by
Aksu-Koç (1998) on child Turkish because the investigation involved the
following four qualities: (1) it was sensitive to the emergence of TAM morph-
ology in a very young child (Deniz 1;3 to 1;11), (2) it included child-directed-
speech, (3) the classification of Aktionsart was empirically motivated, and
(4) the results revealed an unambiguous link to Aktionsart. Four multifunc-
tional affixes of Turkish were analysed, each having a value of tense, aspect
and modality (i.e. specifically evidential). The affixes with their meanings
and examples are as follows: (1) -dI past, perfective, and direct experience,
gel-dI ‘he came’; (2) -Iyor present and progressive, gel-Iyor ‘he is coming’; (3) -Ir
habitual-generic and possibility, gel-Ir ‘he comes’; and (4) -mIş past/present,
resultative, and indirect experience, gel-mIş ‘he has come evidently’. A set of
tests motivated by Dowty (1979) and organized for child language research
by Shirai and Andersen (1995) were used to categorize Deniz’s verbs into
categories of Aktionsart (i.e. state, activity, achievement and accomplish-
ment). Aksu-Koç (1998: 263 and 264) identified four stages of development as
follows:
1. 1;5.9 to 1;9.9: emergence and restricted use of -dI with achievement verbs
(e.g. bul- ‘find’)
2. 1;7.3 to 1;7.8: flexible use of -dI and emergence and restricted use of -Iyor
with activity (e.g. ye- ‘eat’) and state verbs (e.g. yat- ‘lie:down’)
3. 1;7.23 to 1;8.14: flexible use of -dI and -Iyor and emergence and restricted
use of -Ir and -mIş
4. 1;8.27 to 1;10.19: flexible use of all four forms.
Table 18.3 The percentage of verb tokens produced with -ta and with -te i- by
Sumihare for two predicate categories and three age periods
(1993: 200 and 201) analysed the emergence of -ta from 0;11 to 1;8 and -te i-
from 1;11 to 2;2. Table 18.3 contains the percentage of verb tokens produced
with -ta and with -te i- by the child Sumihare during three particularly salient
acquisition periods. For -te i-, the interval of 2;2 was split in half due to the
rapid development at age 2;2. The results show the relationship between
elements of predicate structure and tense-aspect morphology. At an early
age, the relatively high percentage of verb tokens were found in achieve-
ments with -ta (i.e. ‘past/perfective’) and activities with -te iþru (i.e. the action-
in-progress meaning). As the developmental process rapidly unfolded, the
likelihood of activity verbs with -ta increased, and the likelihood of achieve-
ment verbs with -te i- (i.e. the resultative state meaning) increased. As was
found with Deniz learning Turkish, the link between the do′ (x) predicate
structure and present-progressive meaning and between INGR pred′ (x) struc-
ture and past-perfective meaning was found in the Japanese child Sumihare.
However, like Aksu-Koç’s (1988) Turkish findings, there was no ‘restrictive’
stage since Sumihare inflected activity verbs as well as achievement verbs
with -ta.
Shirai (1998) continued his investigation of the relationship between
tense-aspect morphology (i.e. -ta, -te i-, and -ru) and categories of Aktionsart
with three more Japanese children: Aki (1;5 to 3;0) (Miyata 1995, cited by
Shirai), Yocchan (1;11 to 2;2) (Clancy 1985), and Taachan (1;10 to 2;2) (Kokur-
itsu, Kokugo and Kenkyujo 1982, cited by Shirai). When the tense-aspect
morphemes became productive, each child had a relatively high frequency
of verb types in one of the Aktionsart categories. For Aki, achievement -ta
verb types were the most frequent at 2;1 and remained highly frequent to
2;7, and activity-te i- verb types emerged as the most frequent at 2;4 and
remained so to 2;7, supporting Shirai’s (1993) investigation of Sumihare.
Aki’s mother’s child-directed speech was also analysed, and in general, the
relationship between tense-aspect morphemes and Aktionsart categories
supported Aki’s acquisition pattern. However, the ‘expected’ pattern was
not complete for Yocchan and Taachan. Following expectations, at product-
ive usage, the achievement category was closely linked to past -ta, but this
was not the case for durative -te i-. Hence, the expectations of Aksu-Koç’s
‘restricted stage’ are not always realized.
Predicate Structure, Operators and Finite Morphology in Polish. Weist
et al. (2004) utilized the corpora of six children learning Polish (1;3– to
4;11) and six children learning English (1;2 to 4;11) from the CHILDES
archives (MacWhinney and Snow 1985) to investigate the emergence of
tense-aspect morphology within the RRG theoretical framework. The
method of investigation was ‘predicate tracking’. Employing Dowty-type
tests, the process began by assigning verbs to four basic Aktionsart categor-
ies (i.e. state, activity, accomplishment and achievement). Following
Aktionsart classification, dynamic predicates were partitioned into telic
(i.e. accomplishment and achievement) and atelic (i.e. activity) sets (see
Table 18.2). Polish has perfective and imperfective aspect where imper-
fective is the least marked form, and English has progressive and non-
progressive aspect where the non-progressive (or ‘simple’) form is the least
marked. In Polish, there is an interaction of tense and aspect such that
non-past perfective verbs have future-tense meaning, while non-past
imperfective verbs have present-tense meaning. Imperfective future has a
periphrastic form. English has quasi-modal (i.e. going-to/will) future forms.
With two values of aspect and three values of tense, there might be six
inflectional forms to evaluate. However, present tense in perfective aspect
does not exist in Polish, and the children learning English in this study did
not use the future tense progressive aspect form. In order to balance the
cross-linguistic comparison, we reduced the statistical analyses to four
tense-aspect forms in each language:
1. Polish:
(a) perfective past and non-past (future meaning)
(b) imperfective past and non-past (present meaning)
2. English:
(a) non-progressive past and (going-to / will) future
(b) progressive past and present.
Table 18.4 The average age of the emergence of tense-aspect forms for
(atelic/telic) predicates
(a)
Polish Aspect
* {} The future imperfective data were scored but could not be compared to English.
(b)
English Aspect
* {} ‘Present’ progressive forms were scored with and {without} the auxiliary.
The Weist et al. (2004) predicate tracking study evaluated three depend-
ent variables. In addition to the age of emergence (see Table 18.4), the
likelihood that a tense-aspect form would occur and contrasts in tense-
aspect forms were also investigated. The likelihood data complement the
emergence data; for example, a form that emerged early was highly likely.
Contrast is an important indicator of acquisition. Children acquiring a
relatively inflected language like Polish do not hear bare stems and neither
do they produce them. A tense and/or aspect morpheme may emerge as a
‘frozen’ form, and contrast provides evidence for functional status. Three
types of contrast were evaluated for all predicates (i.e. telic and atelic
combined) (1) tense only (e.g. past perfective / future perfective), (2) aspect
alone (e.g. past perfective / past imperfective), and (3) tense-aspect (e.g. past
perfective / present imperfective). The contrast findings are presented in
Table 18.5. Contrary to conventional wisdom, there was no evidence to
support the claim that aspect is acquired before tense. Furthermore, there
was considerable evidence that tense morphology codes deictic relations.
Weist et al. (2009) conducted a second predicate tracking study in order to
investigate the impact of predicate structure on the acquisition of
agreement and tense morphology. The data for this study were found in
the corpora of six children learning Polish (1;0 to 4;11) and six children
learning English (1;2 to 4;11) taken from a variety of sources (see, for
example, the Kraków project, Smoczyńska 1985). The set of predicates being
tracked were either stative or dynamic, with the dynamic set further
classified as atelic versus telic. The telic subset included predicates with
Table 18.5 The average age of the emergence of tense-aspect, tense, and aspect
contrasts summing over values of lexical aspect
Table 18.6 The percentage of agreement and tense contrasts for the three predicate
types in Polish and English
Type Agreement
Type Tense
imperfective aspect, and (3) making reference to remote prior events. With
deference to Piaget’s theory of conceptual development (Piaget 1954, 1971),
the child language researchers ruled out the deictic function of past-tense
morphology, for example, ‘the child lacks an abstract conception of time
that would allow him to construct, between any two events X and Y, the
relation Event Y precedes Event X, even when no other type of relation holds
between event X and Y’ (Antinucci and Miller 1976: 184). Combining longi-
tudinal observations with experimental findings, Weist et al. (1984) dis-
covered that none of the three claims about the emergence of past-tense
morphemes were correct for child Polish. During spontaneous caregiver–
child interactions and within controlled elicitation conditions, 2-year-old
Polish children produced activity verbs inflected for imperfective aspect in
past tense referring to moderately remote prior events. In fact, the Polish
findings are supported by the Turkish (Aksu-Koç 1988) and Japanese (Shirai
1998) data, revealing a low frequency of activity predicates inflected for past
tense (see Table 18.3 for Japanese). Regarding the Piagetian conceptual
limitation argument, decades of research with infants and toddlers has
demonstrated that young children have the capacity to remember past
experiences (see, for example, Bauer 1996, 2007 and Rovee-Collier 1997).
The idea that the concept of aspect is functional in child language prior to
the concept of tense may also be influenced by a methodological artefact.
Child language data have often been collected in a ‘here-and-now’ restricted
environment; see, for example, Aksu-Koç (1998: 261) ‘During the sessions
Deniz and her mother played with toys, read, looked at pictures in books,
and talked.’ In contrast, Bowerman (1981: 1 and 2) took ‘copious notes’ on
her daughter’s utterances throughout the day, and she found that, ‘Spon-
taneous reference to both past and future (and nonactual) events was well
established from the one-word period.’ Some of Bowerman’s observations of
Christy’s temporal utterances are found here in Table 18.7. Christy’s capacity
to talk about the ‘not-here-and-now’ is quite consistent with the develop-
mental research on memory processes. In other words, children who can
think about prior and subsequent events can also talk about these events if
given the opportunity. Table 18.7 shows that when Christy was producing
uninflected two- and three-word combinations, she was locating event time
prior-to, at, and subsequent-to speech time. As the inflectional morphology
emerged, her use of tense matched the utterance context. Regarding past
time utterances, the Aktionsart of her verbs was atelic as well as telic.
Relating to the ‘layer-emergence’ hypothesis, while wider-scope operators,
such as tense, incur greater linguistic complexity, the more recent develop-
mental findings indicate that young children have the capacity to process
this information (see also Table 18.5).
‘Operator-Scope’ Hypothesis: Temporal Domain. Within the RRG frame-
work, the concept of scope defines the relationship between operators and
layers of clause structure. The ‘operator-scope’ hypothesis predicts that
children understand scope relations as the TAM morphology emerges.
Past (1;5). Write Sissy. She notices where her sitter had written on her hand.
Present (1;6) Kiddicar ride. Said while playing with the kiddicar.
Future (1;9) Pie open later. She anticipates taking an unopened pie home.
Table 18.8 The initial occurrence of the temporal adverbs in three sets of adverbs
* The initial occurrence of these adverbs always involved at least one tense-adverb
deictic contradiction.
(I) had (a) pontoon and already it broke’, the adverb jutro is pointing to the
future, and the tense (-ł-) is pointing to the past. From the perspective of
cognitive development, Friedman (1978: 281) argued that such incongruities
signalled a deficiency in the child’s early tense system. However, to the
contrary, the Polish data showed that children initially coordinate the
‘immediate’ temporal adverbs (już ‘already’, teraz ‘now’, and zaraz ‘soon’)
with tense. The tense–adverb incongruities emerge at a later phase of
acquisition when children are learning the meaning of the ‘cyclic’ adverbs.
The child’s earlier understanding of tense remains intact. While a direct
comparison of the emergence of temporal adverbs with tenses was beyond
the scope of the Weist and Buczowska study, the Polish data cited in
Table 18.4 indicate a close proximity. Relating to the ‘operator scope’
hypothesis, the Polish data support the claim that children demonstrate
an understanding of different layers of structure when tense-aspect morph-
ology is acquired.3 The aspectual adverb już ‘already’ modifies the nuclear
layer and the temporal adverbs teraz ‘now’ and zaraz ‘soon’ (not to mention
the cyclic and remote adverbs) modify the core layer.
‘Operator-Scope’ versus ‘Layer-Emergence’ Hypothesis: Modal Domain. In
the temporal domain, the ‘layer-emergence’ argument proposes that mor-
phemes with the potential to code deictic relations (i.e. tense) in the adult’s
language code temporal contour relations (i.e. aspect) in the child’s language.
Within the RRG framework, this argument is motivated by the fact that
aspect modifies a more primitive layer of clause structure than tense. There
is a companion argument in the modal domain (Van Valin 1991: 17, Fig. 3).
Since the operator ‘modality’ (covering deontic modality) is a core operator
and status (including epistemic modality) is a clause operator, along with
evidentials, deontic modality should be acquired before epistemic or eviden-
tial modality. Considering data from children learning Greek and English,
Stephany (1986) presented the argument that ‘Epistemic modal meanings
develop later than deontic ones in language acquisition’ (p. 393) and further-
more, the reason for this acquisition sequence can be found in the Piagetian
view of cognitive development. According to Stephany (1986: 393), ‘Studies of
Table 18.10 Typology of restricted neutralization of semantic roles (Van Valin and
LaPolla 1997: 269)
Acehnese no no No Semantic
Warlpiri yes no Yes [S,A]
English yes yes Yes [S,A,d-S]
Dyirbal yes yes Yes [S,U,d-S]
Table 18.11 Average age and range of initial person agreement contrasts in English
and Polish
English Polish
acquisition process is more rapid in Polish, the more highly inflected language.
Limiting the analysis to Polish, we investigated the children’s capacity to
transfer their knowledge of agreement from one tense to a second tense. We
calculated the number of months it would take a child to make the transfer (i.e.
lag time), and we found that for some of their predicates, all of the children
transferred their capacity to form contrasts in agreement from one tense to
another within one month (i.e. zero lag). By approximately 2;4, restricted
neutralization is a property of child Polish, but exactly how do we classify
child Polish at this phase of development? At 2;4, it appears that child Polish
should be classified as an [S,A] type like Warlpiri at least. However, like English,
Polish has passive voice, and the Weist, et al. (2009) study did not evaluate this
possibility. This possibility was the focus of a paper by Weist (1990).
Weist (1990) investigated all of the facets of restricted neutralization in
child Polish, including the concept of voice. The research included the cor-
pora of four Polish children with ages ranging from 1;7 to 1;11 for two
children and 2;1 to 2;6 for the other two children. The children’s TAM system
was productive with contrasts in tense, aspect and mood. Transitive verbs
were sometimes observed with a direct object, and intransitive verbs never
were. The children contrasted nominative with accusative case, and their
verbs agreed with the privileged argument in person, number and gender.
During the period from 1;6 to 2;0, the children produced intransitive verbs
with a theme or patient as the core argument (i.e. with the macrorole Under-
goer) (e.g. spasˊ ć ‘to fall down’, przyjsˊ ć ‘to come/arrive’, and uderzyć si˛e ‘to get
hit’), and they produced intransitive verbs with an agent core argument (i.e.
with the Actor macrorole) (e.g. ˊspiewać ‘to sing’, płwać ‘to swim’, and bawić si˛e
‘to play’). Regarding the pseudo-reflexive particle si˛e (refl), the children
followed the rules for si˛e placement as follows: (1) immediately after the
verb: Marta (1;8) kr˛eci si˛e ‘(it) turns’, and (2) optionally, after the first stressed
element in the sentence: Marta (1;10) co si˛e kr˛eci? ‘what is turning?’ versus
Marta (1;9) czemu kr˛eca si˛e kóleczka? ‘why are the little wheels turning?’. In
Polish (child as well as adult), the transitive/intransitive contrast is distinct-
ive, as the following observation demonstrates: Kubusˊ (2;4) Otwiea-m. Tak si˛e
otwiera-Ø ‘(I) am opening (1sg) (the box). (It) opens (3sg) like this’. Here it is
clear that Kubusˊ has shifted from the transitive verb otwierać to the intransi-
tive otwierać si˛e. The transitive verb otwierać is inflected for first-person
singular, the pronoun is dropped, and the direct object is omitted. The
particle si˛e with the intransitive verb otwierać si˛e has been moved to the
position after the word tak, and the verb is inflected for third singular
agreeing with ‘it/box’. In summary, there is clear evidence children have
the capacity to neutralize the semantic roles Undergoer and Actor for
syntactic purposes. However, these observations demonstrate neutralization
that is restricted to the intransitive context. Polish children produce passive
participle verb forms in adjectival constructions (e.g. Bartosz (1;11) Zepsute, to
zepsute jest ‘Broken, (it) is broken’), but they do not produce passive voice
constructions (see Weist 1990: 1338–1341). Hence, child Polish has the prop-
erties of an [S,A] language (like Warlpiri) and not the fluent [S,A, d-S] lan-
guage (like adult Polish).
Linking Rules. Does child language contain grammatical relations, and if
so, what kind? While there are variations on these themes, there have been
two major approaches to answering this question. One approach begins with
the investigation of child language data and derives a hypothesis about
grammatical relations from what children produce and comprehend (see,
for example, Brown 1973; Bowerman 1973). The second approach begins
with a linguistic theory and makes predictions about what should be found
in the child language data that correspond to that theory (see, for example,
Pinker 1984; Radford 1990). Invariably, the investigators that begin with the
child language data arrive at the conclusion that grammatical relations are
learned, and those that begin with a linguistic theory (for example, some
version of Chomsky’s thinking) assume at the outset that grammatical
relations are innate. A revealing demonstration of these two alternatives
concerns Pinker’s concept of innate linking rules, and Bowerman’s observa-
tions of language acquisition in her daughters Christy and Eva. According to
Pinker (1984), children are innately equipped with linking rules that initiate
the process of mapping thematic roles onto syntactic functions. A hierarchy
of thematic roles and a hierarchy of syntactic functions were proposed to
establish the linking process. The hierarchy for thematic roles contains
agent, theme/patient, and location/goal/source, and the synchronized hier-
archy for syntactic functions contains subject, object and oblique object, in
that order. Direct links between thematic roles and syntactic functions (e.g.
agent to subject and patient to object) are ‘canonical’ (i.e. the innate
default), and the theory predicts that learning will be facilitated. The set
of thematic relations is defined by verb-argument structure. For an activity
predicate like to cry, the logical structure contains a single agent core
argument, and the agent is canonically linked to subject. An achievement
predicate like to fall has a single theme core argument, and since there is no
agent in the predicate structure, the theme is canonically linked to subject.
Non-canonical mapping occurs when the default links are ‘crossed’; for
example, the predicate to have contains the core arguments of location and
theme, where the location role becomes subject and not the higher-ranking
role of theme. For verbs taking three arguments such as to give, the canonical
Two-argument verbs
Three-argument verbs
Child Age Canonical: Agent – verb – Non-canonical: Agent – verb – goal – theme
theme – goal
18.8 Conclusion
In his paper linking RRG to language acquisition, Van Valin (1991: 9) pre-
sented two ‘perspectives’: (1) ‘Syntax is . . . relatively motivated by semantic,
pragmatic, and cognitive concerns’ and (2) ‘the child actually learns language
and CONSTRUCTS a grammar during the process of language acquisition’.
Regarding how the construction process functions, Van Valin cited Slobin’s
(1985) theory of Operating Principles. According to Slobin (1985: 1158), chil-
dren are innately equipped with a ‘language-making capacity [LMC]’ and
Slobin proposed that ‘LMC must begin life with some initial procedures for
perceiving, storing, and analyzing linguistic experience, and for making use
of capacities and accumulating knowledge for producing and interpreting
utterances’. The theory of Operating Principles has been criticized by child
language researchers, for example Bowerman (1985: 1281), and linguists, for
example Pinker (1989: 463), since Slobin’s information processing theory
lacks a model of grammar. This chapter was designed to investigate the
potential that RRG might be able to fill this void.
Semantic functions such as agent and theme/patient are operational in
the earliest phase of language acquisition, when single-word utterances
dominate the child’s corpus. The child’s capacity to process the finite
morphology of the target language becomes apparent as multiple-
morpheme utterances emerge, especially in highly inflected languages like
Turkish and Polish. As soon as there is evidence for the acquisition of the
elements of a tense-aspect-modality system, the semantic structure of the
predicate influences the pattern of acquisition. Hence, on the one hand,
children are utilizing their innate information processing capacity (which
might be referred to as their LMC), and on the other hand, this capacity to
process information is combined with at least some knowledge of predi-
cate structure (see Tables 18.3 and 18.4). While these two components of
linguistic knowledge are necessary to explain the acquisition data, they are
not sufficient. Children integrate this knowledge within a concept of
clause structure (see Table 18.1). Potential explanations of language acqui-
sition based exclusively on principles of information processing (e.g. Slobin
1985), or with categories of predicate structure (e.g. Shirai 1998) will fall
short if they fail to explain how integration into clause structure is accom-
plished. From the perspective of RRG, clause structure is created from the
semantic principle of predicate-argument structure, and it contains levels
of semantic complexity. The ‘operator-scope’ hypothesis presented here
predicts that children identify the scope of operators as the concepts of
tense, aspect, and modality are being acquired. The predicate tracking data
(e.g. Weist et al. 2004) showed that Polish children acquire tense (i.e. a
clause operator) and aspect (i.e. a nuclear operator) during the same phase
of acquisition. Similarly, Choi’s study of the emergence of evidential mor-
phemes (i.e. clause operators) found early productivity of clause operators
in the modal domain of the TAM system. These facts require that children
are constructing their understanding of clause structure simultaneously
with their decoding of the elements of the tense-aspect-modality system.
Van Valin’s (2002) study of the acquisition of subject–auxiliary inversion
revealed how the integration of the tense operator within clause structure
explains the emergence of wh-questions in child language data. From the
viewpoint of cognitive development, the child language data do not sup-
port the idea that the infant-toddler is somehow limited in their capacity
to process complex relations and construct meaningful representations in
the manner imagined by some early child language researchers such as
Braine (1976) or more recent researchers such as Rowland and Pine (2000,
2003). In contrast, the child language data reviewed here are consistent
with Spelke’s ‘core knowledge systems’ theory of cognitive development,
where the core knowledge systems are thought of as ‘mechanisms for
representing and reasoning about particular kinds of ecologically import-
ant entities and events’ (Spelke 2000: 1233). In retrospect, Slobin’s (1985)
set of Operating Principles might be thought of as the core knowledge
system for language.
References
Aksu-Koç, Ayhan. 1988. The Acquisition of Aspect and Modality: The Case of Past
Reference in Turkish. Cambridge: Cambridge University Press.
Aksu-Koç, Ayhan. 1998. The role of input vs. universal predispositions in the
emergence of the tense-aspect morphology: Evidence from Turkish. First
Language 18: 255–280.
Antinucci, Francesco and Ruth Miller. 1976. How children talk about what
happened. Journal of Child Language 3: 167–190.
Bauer, Patricia. J. 1996. What do infants recall of their lives? Memory for
specific events by one- to two-year-olds. American Psychologist 51: 29–41.
Bauer, Patricia. J. 2007. Remembering the Times of Our Lives: Memory in Infancy
and Beyond. Mahwah, NJ: Lawrence Erlbaum.
Bowerman, Melissa. 1973. Early Syntactic Development. Cambridge: Cambridge
University Press.
Bowerman, Melissa. 1981. Notes for the Nijmegen Workshop. MPI, Nijmegen,
Netherlands.
Bowerman, Melissa. 1985. What shapes children’s grammars? In Dan I.
Slobin (ed.), The Crosslinguistic Study of Language Acquisition, Vol. 2,
1257–1319. Hillsdale, NJ: Lawrence Erlbaum.
Notes
Abbreviations
19.1 Introduction
*
I would like to thank Delia Bentley, Evelina Fedorenko, David Kemmerer, Jean-Pierre Koenig and Anja Latrouite for helpful
discussion and comments on earlier drafts. Some of this research was supported by the German Research Foundation (DFG)
through Cooperative Research Center 991 ‘The structure of representations in language, cognition and science’ at the
Heinrich Heine University Düsseldorf.
Baynes and Gazzaniga (1988: 124) report that two split-brain patients show
the same linguistic ability out of their right hemispheres that the ‘agram-
matic’ aphasics in Linebarger et al. (1983) exhibited: ‘The most striking
finding in these two commissurotomized patients with right hemisphere
language is the dissociation between the [impaired-RVV] ability to compre-
hend syntactically constrained sentences and the [intact-RVV] ability to
judge their grammaticality. This dissociation can be found in two distinct
populations, agrammatic aphasics and commissurotomy patients.’ These
results will be discussed in detail in Section 19.5, and an explanation for
them will be proposed.
How, then, is it possible for an isolated right hemisphere to make gram-
maticality judgements? If the rules, representations, constraints, etc., that
Until the early 1990s the view that the primary language areas of the brain
are Broca’s and Wernicke’s areas, which are in the left frontal and temporal
lobes, respectively, and which are connected by the arcuate fasciculus, was
widely assumed to be basically correct. Broca’s area includes the pars oper-
cularis and the pars triangularis in the inferior frontal gyrus, while Wer-
nicke’s area involves the superior temporal gyrus and the angular gyrus. The
locations of these two areas are represented in Figure 19.1.
However, evidence that other parts of the brain were directly involved in
language processing began to accumulate. Mazoyer et al. (1993) showed that
when French speakers were asked to listen to well-formed French sentences,
a list of words, and Urdu sentences, there was activation in the anterior
temporal lobe only when French sentences were heard, not in any other
condition. Of particular relevance for this discussion is the claim put forth
in Dronkers et al. (1994) that the anterior part of Brodmann’s area [BA] 222 in
the left temporal lobe is involved in syntactic processing based on a study of
templates with thematic relations, the RRG templates would have the
semantic macroroles actor (A) and undergoer (U). Hence the default canon-
ical sentence template for English would be RP:A-NUC-RP:U. The advantage
of using macroroles can be seen in (1).
(main) clause in English; the exact values of these operators are a function of
the clause being processed. There is also an optional place for peripheral
adjuncts; only the core-level periphery is shown here, but there may be periph-
eries modifying the nucleus, core or clause. In Figures 19.3 and 19.4, the two
basic intransitive templates are given. The first is for verbal nuclei, and the
(2) a. Mary presented the first-place trophy to Sam. Oblique core argument
a0 . Mary presented Sam with the first-place trophy. Oblique core argument
b. Sam put the first-place trophy in a box /on the shelf. Oblique argument-adjunct
c. Sam gave the box to his assistant. Oblique core argument
The two things that are most important for the rapid interpretation of a
template is the assignment of semantic macroroles and non-predicative
prepositions. For a simple transitive verb like kill or smash, the logical
structures [LSs] would be as in (3).
(3) a. General: [do0 (A: x, [affect0 (x, y)])] CAUSE [INGR pred0 (U: y)]
b. kill: [do0 (A: x, [affect0 (x, y)])] CAUSE [INGR dead0 (U: y)]
c. smash: [do0 (A: x, [affect0 (x, y)])] CAUSE [INGR smashed0 (U: y)]
Once the template in Figure 19.2 has been matched and the verb recognized,
the basic ‘who did what to whom’ meaning is available by integrating the
arguments from the template with the LS. This can be illustrated with The
burglar smashed the window.
(4) present: a. [do0 (A: x, [affect0 (x, z)])] CAUSE [INGR have0 (NMR: to y, U: z)]
b. [do0 (A: x, [affect0 (x, z)])] CAUSE [INGR have0 (U: y, NMR: with z)]
Once the template in Figure 19.5 has been matched and the verb recognized,
the basic ‘who did what to whom’ meaning is available by integrating the
arguments from the template with the LS. For verbs like give that occur in
the ‘double-object’ construction (e.g. Sam gave his assistant the box), their LSs
would be as in (5).
(5) give: a. [do0 (A: x, [affect0 (x, z)])] CAUSE [INGR have0 (NMR: to y, U: z)]
b. [do0 (A: x, [affect0 (x, z)])] CAUSE [INGR have0 (U: y, NMR: z)]
The LS in (5a) integrates with the template in Figure 19.5, while the one in
(5b) integrates with the template in Figure 19.6.
The comprehension of a passive sentence such as The window was smashed
by the burglar is just as straightforward.
2. The lexical entry for the predicate in the nucleus in (5a) is selected because
of the NMR: to specification and is activated, and the annotations on the
arguments in the template and the LS guide the integration (i.e. RP: A:
Mary ¼ A: x, ∴ Mary ¼ x, etc.), yielding the interpretation
[do0 (A: Mary, [affect0 (. . .)])] CAUSE [INGR have0 (NMR: Sam, U: what)].
This template works for a variety of verbs but not all (see Shimojo, Chap-
ter 16): What did Mary think/believe/expect/anticipate/claim/deny/doubt/assert that
Sally gave to Sam? These templates directly address the interpretation of the
wh-expression and do not put it off until the rest of the sentence has been
processed, as the linking algorithm does.
The template in Figure 19.11 introduces templates for complex sentences,
and in this case the structure involves subordination (embedding). There are
non-embedded complex sentences as well, and sentences like Mary persuaded
Sam to wash the dishes are represented by templates like the one in
Figure 19.12.
Sentences with this structure (e.g. Mary persuaded/promised Sam to wash the
dishes) obligatorily ‘share’ an argument; that is, the infinitival complement
is missing an argument, and accordingly one of the core arguments in the
first core is interpreted as also filling a role in the second core. In this
sentence the actor of wash is missing; when the verb in the first core is
persuade, the undergoer Sam is interpreted as the actor of wash, whereas
when the verb is promise, the actor Mary is so interpreted. The LS for the
persuade version would be [do0 (Mary, [affect0 (Mary, Sam)])] CAUSE [want0
(Sam, [do0 (Sam, [wash0 (Sam, dishes)])])]. Despite multiple occurrences in the
LS, the RP Sam appears only once in the sentence as the undergoer of
persuade. These last two examples barely scratch the surface of the intricacies
of complex sentences, but they illustrate how the system can be applied
to them.
What happens when breakdowns such as garden paths occur? What
happens when an unfamiliar structure is encountered? These are situations
in which step-by-step reanalysis and reinterpretation is required. On top of
the template-based processing discussed earlier, there is also the possibility
of step-by-step processing as well. In this mode, the syntactic representation
of the sentence is constructed, and then the steps of the syntax-to-semantics
linking algorithm are carried out literally, which permits double-checking
of all relevant aspects of the construction.
All of the examples of linking with semantically augmented syntactic
templates have been from English, which has a relatively rigid word order
that is crucial for the interpretation of the core arguments. Can this
approach to language comprehension be applied to languages with various
degrees of word-order flexibility and which rely on case marking rather than
linear order for the interpretation of core arguments? The answer is ‘yes’,
and the application to case-marking languages with very flexible word order
will be illustrated with an example from Croatian.
(9) darovati ‘give as a gift’ [do0 (A: x, [affect0 (x, z)])] CAUSE [INGR have0 (NMR: y, U: z)]
a. Unuc-i su bak-i darova-l-i cvijeć-e.
grandson-m.pl.nom be.3pl grandmother-f.sg.dat give-pst-pl flower-m.pl.acc
‘The grandsons [x] gave flowers [z] to [their] grandmother [y].’
b. Baki su unci darovali cvijeće.
c. Cvijeće su baki unci darovali.
d. Baki su darovali cvijeće unci.
As in (2a0 ) and the LS in (4b), but not in (9), the y-argument in the LS is the
undergoer, and the z-argument is the non-macrorole core argument in the
instrumental case, which is analogous to the instrumental preposition with
in English. As in (9), all of the examples in (10) have the same meaning. The
crucial difference between (9) and (10) is the coding of the non-macrorole
argument, dative in (9) and instrumental in (10), and accordingly it must be
specified in the LSs for darovati, as in (11), just like in (4) for English present.
It is clear from the examples in (9) and (10) that linear order plays no role
in the interpretation of these sentences and that the interpretation of the
core arguments is based on case marking. In Van Valin (2006) ‘case associ-
ation principles’ were used to handle the linking between case forms and
macroroles. The grammar of Croatian includes the ones in (12) as well as
additional ones for cases which are not found in (9) and (10).
The nominative case codes the actor with some intransitive verbs (those
with an activity predicate in their LS) and active-voice transitive and ditran-
sitive verbs, and it codes the undergoer with some intransitive predicates
(those without an activity predicate in their LS) and with passive-voice
transitive and ditransitive verbs. The accusative case marks the undergoer
of an active-voice transitive or ditransitive verb. The other two cases mark
non-macrorole arguments, dative arguments being direct core arguments,
while instrumental ones are oblique core arguments.
In RRG, whether an XP is a core argument or a peripheral adjunct is a
consequence of the semantic function of the XP and not on its place in the
linear string. (Languages may impose linearization restrictions, of course.)
This means that the syntactic templates which represent the layered struc-
ture of the clause need not have a fixed order, and this is what languages
like Croatian require: unordered templates. The only ordering constraint in
the clause concerns the auxiliary, which must be the second word in all
clauses. This can be represented as in Figures 19.13 and 19.14.
The dotted lines connecting the RPs and the nucleus to the core node
indicate that the order of these elements is not fixed but that any of
them could occupy any of the positions. The solid line connecting the
AUX node to the core node signals, on the other hand, that its position
in the core is fixed and it must invariably be in the second position.
Information structural considerations like the relative topicality vs.
focality of the referents of the RPs strongly influences the order of words
in context, as is well known. Croatian not only allows the RPs and
nucleus to occur in any order, but it also permits discontinuous constitu-
ents, as in (13a) vs. (13b).
Je ‘is’ is like su ‘are’ in (9) and (10), as it must be second in the clause, and
second position may be defined as being after the first constituent, as in
(13a), or after the first word, as in (13b). (Native speakers prefer (13b).)
Determiners are connected to the NUCLEUSR via the operator projection of
the RP, and Croatian, unlike English, does not require strict adjacency
among RP constituents, nor is their order fixed. Hence in Croatian linking
templates like those in Figures 19.13 and 19.14, not only are the constituents
in the clause unordered, but there are few constraints on the RP templates
as well.12 These templates apply, accordingly, to all of the word-order vari-
ants in (9) and (10), as well as others not exemplified, and other three-
argument verbs with a dative or instrumental third argument. The match-
ing process goes as follows for (9c), for example.
To sum up, the determination of the syntactic structure and the semantic
interpretation of an utterance is achieved rapidly due to the annotations on
both the templates and the lexical representations of predicates. This is
carried out in BA 22 using both anterior and posterior resources, the ‘sen-
tential combinatorics’ mentioned in the previous section, as well as other
areas of the left temporal lobe. What was called ‘early structure building’ in
the previous section is, on this view, the assembling of syntactic templates,
for instance selecting the RP templates for the RP slots in the clausal
template, and the matching of the resulting template with the utterance
being processed. The integration of the information in the lexical represen-
tation of the predicate in the nucleus of the clause yields a basic ‘who did
what to whom’ meaning (a ‘good enough’ interpretation). In the case of
breakdowns, the system can execute the syntax-to-semantics linking algo-
rithm in a controlled, step-by-step fashion.
At the end of Section 19.3, the following question was raised: What does
BA 22 process, such that it appears to have syntactic properties which the
anterior part handles and semantic properties which both parts handle? The
answer should now be clear: semantic-macrorole-augmented syntactic
linking templates and the lexical entries for predicates. Given the integra-
tion of both syntactic and semantic information in the linking templates
and in the lexical entries for predicates, it is in fact difficult to claim that
anterior BA 22 is primarily for syntactic processing and posterior BA 22 is for
semantic processing. Rather, all of BA 22 is involved both aspects of lan-
guage comprehension, consonant with the point made by Federenko et al.
(2020) mentioned earlier. They are also crucial for the resolution of the
mystery stated at the end of Section 19.2: How is it possible for the isolated
right hemisphere in a split-brain patient to make grammaticality
judgements?
the linking template for that construction, which would share structural
and semantic macrorole properties with non-reflexive constructions (e.g. Bill
asked Sally to help Mary).14 Like lexical items, these templates are stored in
both hemispheres. The second kind of grammatical resource consists of the
mechanisms for selecting and interpreting templates; they are primarily in
the left temporal lobe, including BA 22. The interpretation of templates
includes the integration of the semantic representations of lexical items
with the templates based on the semantic role annotations and other infor-
mation, as well as the evaluation of the sentence with respect to any
constraints on the construction. Thus what may be termed ‘comprehension
competence’ is distributed: it is found in the content of the linking tem-
plates, in the content of the lexical representations of verbs and other
predicates, and in the interpretive mechanisms.
After the commissurotomy, what are the linguistic resources in the right
hemisphere? At most there would be lexical items in the lexicon and linking
templates in the LTI. They are cut off from the left-hemisphere interpretive
mechanisms mentioned earlier. Yet a few commissurotomized patients are
able to make grammaticality judgements using their right hemisphere. The
main grammatical information available in their right hemisphere is in the
linking templates, namely, the structure of phrases and clauses, and there is
minimal grammatical information in the lexical entries for verbs: while the
LSs cannot be interpreted semantically, the number of arguments and the
preposition(s) marking oblique arguments can be read off the LSs. Conse-
quently these two sources of information must be the basis for the gram-
maticality judgements. The issue, then, is how is this grammatical
information sufficient for making grammaticality judgements but insuffi-
cient for comprehending ‘syntactically constrained sentences’ like revers-
ible passives, as discussed in Section 19.2.
The linking templates contain three kinds of information: a representa-
tion of phrase structure, the semantic roles of the arguments in the
structure, and construction-specific constraints. The semantic role infor-
mation and the construction-specific constraints (e.g. (14)) cannot be pro-
cessed in the right hemisphere, and so they are in effect invisible. The two
kinds of information express features overlaid on the phrase structure, and
they require the sophisticated neural mechanisms found in the left tem-
poral lobe for their interpretation. This accounts for the inability to cor-
rectly interpret reversible passives. The default strategy of interpreting the
first argument as the actor is a manifestation of the ‘subject preference’
(Wang et al. 2009), which has been found to be operative across languages,
and not from an ability to interpret some of the semantic role annotations
on the templates but not others. It is primarily the phrase-structure infor-
mation that makes possible the grammaticality judgements through the
mechanism of pattern matching, something the right hemisphere excels
at. It is heavily involved in face recognition, for example, and has its own
pattern-matching resources independent of the left hemisphere, unlike
J.W.’s right hemisphere performed surprisingly well, and the three tests
with the biggest difference in performance between the two hemispheres
were the three non-phrase-structural tests involving agreement and/or co-
indexing. Baynes and Gazzaniga (1988) comment on J.W.’s performance:
The other patient discussed, V.P., had nearly identical performance in both
hemispheres.
The results themselves, however, and their implications for the status of
syntactic knowledge in agrammatism, are quite clear: the
comprehension deficit in agrammatism does not reflect loss of the
capacity to analyse syntactic structure. We are left now with the
important question of why agrammatic aphasics seem unable to make
use of this capacity in the comprehension and production of sentences.
(Linebarger et al. 1983: 390)
Linebarger et al.’s subjects have mean A0 values of .94, .91, .91 and .84, which
yields an overall value of .90. The best-performing subject is comparable to
J.W.’s left hemisphere and near the normal controls in Wulfeck’s study,
whereas the worst performer is comparable to the aphasics in Wulfeck’s
study and J.W’s right hemisphere. The similarities in the performance of
J.W.’s right hemisphere and that of the aphasics in the two studies argue
against a solution to the aphasic’s abilities which necessarily excludes J.W.
from consideration.
What about the alternative conclusion, namely, that the aphasics use the
same mechanism as J.W.’s right hemisphere? The explanation given for
J.W.’s behaviour applies equally to the aphasics, and it resolves the paradox
inherent in Linebarger et al.’s interpretation of their results, namely, why is
an aphasic’s grammatical competence accessible for making grammaticality
judgements but not other tasks? The answer is that grammatical knowledge
in the form of linking templates is preserved, but the processing mechan-
isms in BA 22 and other parts of the left hemisphere are out of commission.
Consequently, the aphasics rely on the intact right hemisphere with its
lexicon, LTI and pattern-matching capacity to make grammaticality judge-
ments, just like J.W.’s right hemisphere.
This explanation is not limited to fixed-word-order languages like English.
For a language like Croatian, the analysis presented here predicts that, just
as in English, the argument structure information of verbs will be preserved
d. Step 4: Select the syntactic template(s) for the sentence following the
template selection principles.
Figure 19.15 Bock and Levelt’s model of grammatical encoding and RRG
There are five steps in the linking algorithm. It was pointed out in Van
Valin (2006: 280–282) that these five steps parallel the process of grammat-
ical encoding in the model of a speaker proposed in Bock and Levelt (1994).
This parallelism between a psycholinguistic model of speech production and
an abstract grammatical theory, as in Figure 19.15, suggests strongly that
the RRG semantics-to-syntax linking system has great potential as the basis
for a production model.
The steps in (18) involve applying rules (‘assign macroroles’, ‘select the
PSA’, ‘assign case’, etc.), selecting a bare syntactic template, and inserting
lexical items into the template. The content of these rules, etc., is the same
information contained in the linking templates and morphosyntactically
augmented LSs in the comprehension system, and this is no accident, of
course. As an alternative to applying these linking steps sequentially, it is
possible to condense the linking into constructing the semantic representa-
tion using the augmented LSs with lexical choices for the argument
specified in them. The linking regularities captured by the AUH are gener-
alizations across LSs; in other words, the generalizations expressed in the
AUH are now meta-generalizations across LSs (as in the meta-grammar of (L)
TAG; see Kallmeyer 2010) rather than being a rule constraining the assign-
ment of macroroles. The LSs in (3) express the basic relationship with an M-
transitive verb, while the LSs in (4), (5) and (11) lead to the meta-
generalization that variable undergoer choice is possible with certain
three-place verbs in certain languages. This also holds true for PSA selec-
tion, adposition and case assignment, etc. Thus, the constraints and rules
as well as the hierarchies that inform them are not invoked as grammat-
ical operations to be applied sequentially, as in (18), but rather are concep-
tualized as abstract meta-generalizations across the linking templates and
the augmented LSs.
One of the daunting challenges facing language processing models, both
of comprehension and production, is accounting for the incremental nature
of processing. This issue was addressed in Section 19.4 for comprehension,
where the solution was to posit parallel template processing, whereby the
number of competing templates is reduced as the system encounters more
and more of the input utterance, until eventually there is a single templatic
structure remaining (for unambiguous sentences). With production, on the
other hand, the issue is that speakers can start an utterance without having
planned it to the end (Bock 1995), and accordingly, the linking in (18) is an
idealization which assumes that the entire utterance is planned out in
advance. In reality, the semantic representation in Figure 19.16 is compat-
ible with the possibilities in (19), among others.
The speaker has to decide whether the main clause is to be active or passive
voice, a decision influenced by the relative topicality of the referents of the
RPs CNN and Max, and whether the temporal PP is to occur in its default
position in the core-level periphery or in initial position, and if initial,
whether it is a frame-setting topic in the pre-detached position or part of
the assertion in the pre-core slot. If the speaker begins with during, then
(19a, d) are eliminated, but the remaining four are still possible. If they
begin with CNN, then only (19a) is possible, and likewise if the utterance
starts with Max, only (19d) is possible. There are different commitment
points in the sentence; if it starts with the PP, the speaker is not yet
committed to the voice of the main clause, whereas if it starts with an
RP, the speaker is committed to the form of the main clause and the
placement of the PP.
This example has assumed that the speaker had decided on the predicator,
but it is not unusual for there to be competing predicators with different
implications in play. An alternative way of describing Max’s encounter with
CNN would be with the verb talk (to) as in (20). (The tense has been changed
to present perfect in both semantic representations.)
The speaker has options for the semantic representation as well as the
syntactic form of the utterance, and for the sake of the discussion, it is
assumed that Max is the topic and so the topic expression, the RP Max,
will be the subject of the main clause. Consequently, the main clause
beginning with Max has. . . can be continued with either been interviewed by
CNN or talked to CNN, reflecting two very different syntactic patterns. The
speaker can begin a sentence before deciding on a final structure for
the utterance.
Because structural templates play a crucial role in the processing system,
the incremental nature of processing is captured by competition among
possible templates which is resolved at critical points during the sentence.
For comprehension, the number of potentially relevant templates is reduced
word by word as the receiver processes the sentence linearly. The result of
this winnowing should ideally be a single structure associated with a single
meaning, and multiple structures and meanings create ambiguity. For
production, the speaker can begin uttering a sentence without having
decided the final form and meaning of it. Here again, there are semantic
representations and structural patterns competing, and the choices made at
critical points are a function of the speaker’s intention to communicate a
particular message.
It has become clear that in order to use an abstract competence grammar
for language processing, both production and comprehension, it is neces-
sary to make changes to the system. In particular, the rules and constraints
that the abstract grammar employs in the mapping between meaning and
form and between form and meaning are precompiled in the syntactic
templates and LSs for verbs and other predicators. These linking templates
and augmented LSs play a central role in both production and comprehen-
sion, affording rapid and direct coding in the case of production and what
native speakers perceive as nearly instantaneous interpretation with respect
to comprehension. There are, accordingly, two versions of RRG: the familiar
abstract grammar in Figure 19.17, and the processing system in Figure 19.18.
The same rules, principles, constraints, etc. are captured in both versions,
albeit differently. In the version in Figure 19.17, they are stated explicitly as
part of the linking algorithm, as in (18), whereas in the version in
Figure 19.18 they are meta-generalizations over the linking templates and
augmented LSs.
The double-headed arrow in Figure 19.18 signals that the linking is bidirec-
tional, which is expected in RRG, and it is annotated with ‘Winnowing’ and
‘Discourse-Pragmatics’. ‘Winnowing’ was discussed above. ‘Discourse-prag-
matics’ indicates that, from an RRG perspective, it is an important factor in
the mapping between form and meaning (see Bentley, Chapter 11, Latrouite
and Van Valin, Chapter 12). The interaction with morphosyntax has been
labelled ‘information structure’ and ‘information packaging’, and the latter
term refers to the packaging of the informational content of the utterance by
the speaker in order to facilitate its apprehension by interlocutors. This
entails that discourse-pragmatics affects the linking to syntax more than
the linking to semantics, because the speaker is making choices regarding
everything from lexical choices to word order which can be influenced by it.
Indeed, in Van Valin and LaPolla (1997: 423–430) it was shown that discourse-
pragmatics can interact with and influence every step of the semantics-to-
syntax linking (see also Sections 12.5–12.7 in Chapter 12). It is not the case
that it influences every step in every language; rather, looking across lan-
guages one finds the influence at a given step in one or more languages.
However, there are no steps in Figure 19.18. In terms of the processing version
of RRG, the influence of discourse-pragmatics can be seen in the lexicon and
the LTI. In the lexicon, it affects the choice of LS for the predicator, the choice
of referring expression(s) to fill the argument slots in the LS(s), and even in
some cases macrorole selection. In the LTI, it can influence the choice of
19.7 Conclusion
explanation for the dissociation turned out to involve the same devices as
the right hemisphere grammaticality judgements. While there were only a
small number of split-brain patients and only a small minority exhibited
any linguistic capacity in their right hemisphere, the dissociation in aphasia
has been documented in numerous studies involving speakers from a var-
iety of languages.
The paper began with a reference to Kaplan and Bresnan’s assertion that
there should be a link between abstract competence grammars and process-
ing models. It has been argued that the mechanisms of RRG can be the basis
for neurolinguistic models of both language comprehension and produc-
tion, and the exact nature of the adaptations required by the processing
models have been articulated explicitly and summarized in Figures 19.17
and 19.18. A distinctive characteristic of the RRG account is the distributed
representation of grammatical knowledge in terms of linking templates and
augmented LSs, which, in terms of the dual-stream models, is fully compat-
ible with them being accessible for the dorsal stream (production) as well as
the ventral stream (comprehension). RRG thus meets the requirement that a
grammatical theory claiming to capture speakers’ linguistic competence
should be able to be used in a testable psycholinguistic, or in this case,
neurolinguistic, processing model.
References
Brennan, Jonathan, Yuval Nir, Uri Hasson, Rafael Malach, David J. Heeger
and Liina Pylkkänen. 2012. Syntactic structure building in the anterior
temporal lobe during natural story listening. Brain and Language 120:
163–173.
Brodmann, Korbinian. 1909. Vergleichende Lokalisationslehre der Grosshirnrinde.
Leipzig: Johann Ambrosius Barth.
Clifton, Charles, Jr. and Lynn Frazier. 1989. Comprehending sentences with
long-distance dependencies. In Greg Carlson and Michael Tanenhaus
(eds.), Linguistic Structure in Language Processing, 273–317. Dordrecht:
Kluwer.
Davis, Anthony and Jean-Pierre Koenig. 2000. Linking as constraints on word
classes in a hierarchical lexicon. Language 76: 56–91.
Dronkers, Nina F., David P. Wilkins, Robert D. Van Valin, Jr., Brenda B.
Redfern and Jeri J. Jaeger. 1994. A reconsideration of the brain areas
involved in the disruption of morphosyntactic comprehension. Brain and
Language 47(3): 461–463.
Dronkers, Nina F., David P. Wilkins, Robert D. Van Valin, Jr., Brenda B.
Redfern and Jeri J. Jaeger. 2004. Lesion analysis of the brain areas involved
in language comprehension. Cognition 92: 145–177.
Federenko, Evelina, Idan A. Blank, Mathew Siegelmann and Zachary Miner-
off. 2020. Lack of selectivity for syntax relative to word meanings through-
out the language network. Cognition 203: 1–19.
Ferreira, Fernanda and Nikole D. Patson. 2007. The ‘good enough’ approach
to language comprehension. Language and Linguistics Compass 1(1–2): 71–83.
Friederici, Angela D. 2009. Pathways to language: Fiber tracts in the human
brain. Trends in Cognitive Sciences 13: 175–181.
Friederici, Angela D. 2012. The cortical language circuit: From auditory
perception to sentence comprehension. Trends in Cognitive Sciences 16:
262–268.
Friederici, Angela D., Martin Meyer and D. Yves von Cramon. 2000. Auditory
language comprehension: An event-related fMRI study on the processing
of syntactic and lexical information. Brain and Language 74: 289–300.
Friederici, Angela D., Shirley-Ann Rüschemeyer, Anja Hahne and Christian J.
Fiebach. 2003. The role of left inferior frontal and superior temporal
cortex in sentence comprehension: Localizing syntactic and semantic
processes. Cerebral Cortex 13: 170–177.
Gazzaniga, Michael S. 1983. Right hemisphere language after brain bisec-
tion: A 20-year perspective. American Psychologist 38: 525–537.
Gazzaniga, Michael S. 2000. Cerebral specialization and interhemispheric
communication: Does the corpus callosum enable the human condition?
Brain 123: 1293–1326.
Gazzaniga, Michael S. 2005. Forty-five years of split-brain research and still
going strong. Nature Reviews Neuroscience 6: 653–659.
Gazzaniga, Michael S. and Steven A. Hillyard. 1971. Language and speech
capacity of the right hemisphere. Neuropsychologia 9: 273–280.
Nolan, Brian and Elke Diedrichsen (eds.). 2013. Linking Constructions into
Functional Linguistics: The Role of Constructions in Grammar. Amsterdam: John
Benjamins.
Obleser, Jonas, Jonas Zimmermann, John Van Meter and Josef P.
Rauschecker. 2007. Multiple stages of auditory speech perception reflected
in event-related fMRI. Cerebral Cortex 17: 2251–2257.
Pylkkänen, Liina. 2019. The neural basis of combinatory syntax and seman-
tics. Science 366: 62–66.
Saur, Dorothee, Björn W. Kreher, Susanne Schnell, Dorothee Kümmerer,
Philipp Kellmeyer, Magnus-Sebastian Vry, et al. 2008. Ventral and dorsal
pathways for language. PNAS USA 105(46): 18035–18040.
Schlesewsky, Matthias and Ina Bornkessel. 2004. On incremental interpret-
ation: Degrees of meaning accessed during sentence comprehension. Lin-
gua 114: 1213–1234.
Schwartz, Myrna F., Eleanor M. Saffran and Oscar S. Marin. 1980. The word
order problem in agrammatism: I. Comprehension. Brain and Language 10:
249–262.
Schwartz, Myrna F., Marcia C. Linebarger, Eleanor M. Saffran and David S.
Pate. 1987. Syntactic transparency and sentence interpretation in aphasia.
Language and Cognitive Processes 2: 85–113.
Shimojo, Mitsuaki. 2008. How missing is the missing verb? The verb-less
numeral quantifier construction in Japanese. In Robert Van Valin (ed.),
Investigations of the Syntax–Semantics–Pragmatics Interface, 285–304. Amster-
dam: John Benjamins.
Shtyrov, Yury, Friedemann Pulvermüller, Risto Näätänen and Risto J. Ilmo-
niemi. 2003. Grammar processing outside the focus of attention: An MEG
study. Journal of Cognitive Neuroscience 15: 1195–1206.
Stowe, L. A. 1985. Parsing wh-constructions: Evidence for on-line gap loca-
tion. Language and Cognitive Processes 1: 227–245.
Stowe, L. A., C. A. J. Broere, A. M. J. Paans, A. A. Wijers, G. Mulder, W.
Vaalburg and F. Zwarts. 1998. Localizing cognitive components of a
complex task: sentence processing and working memory. Neuroreport 9:
2995–2999.
Stromswold, Karin, David Caplan, Nathaniel Alpert and Scott Rauch. 1996.
Localization of syntactic comprehension by positron emission tomog-
raphy. Brain and Language 52: 452–473.
Townsend, David J. and Thomas G. Bever. 2001. Sentence Comprehension: The
Integration of Habits and Rules. Cambridge, MA: MIT Press.
Traxler, M. and M. Pickering. 1996. Plausibility and the processing of
unbounded dependencies: An eye-tracking study. Journal of Memory and
Language 35: 454–475.
Turken, And U. and Nina F. Dronkers. 2011. The neural architecture of the
language comprehension network: Converging evidence from lesion and
connectivity analyses. Frontiers in Systems Neuroscience 5: 1.
Notes
Abbreviations
20.1 Introduction
* The research presented in this chapter was supported by the Collaborative Research Centre 991 ‘The Structure of
Representations in Language, Cognition, and Science’ funded by the German Research Foundation (DFG) and the project
TreeGraSP funded by the European Research Council (ERC).
(a) (b)
SENTENCE SENTENCE
CLAUSE CLAUSE
PrCS CORE PERIPHERY CLAUSE
besides the category labels. The second task is to define the modes by which
the syntactic representations are composed from the members of the syn-
tactic inventory (Section 20.3). Next, the templates available in the inventory
must be specified, and this should be done in a way that allows us to capture
generalizations among them within and across languages (Section 20.4).
A further point that needs special treatment is the formalization of the
operator projection (Section 20.5).
A key component of RRG’s approach to syntactic analysis is the layered
structure of the clause: sentences are assumed to have an internal structural
layering consisting of clause, core and nucleus. The different layers serve as
attachment sites for different types of operators: tense operators attach to
the clause, modality to the core, aspect to the nucleus, etc. The core level is
also the default attachment site for arguments. In the following, we will
refer to the subtree of a syntactic representation consisting of the root and
its non-peripheral clause, core, nucleus and predicating descendants as the
clausal skeleton of the representation. The syntactic structures in RRG are
basically labelled trees, and there are good reasons to use tree structures in a
formalization as well. Trees provide the most natural way to analyse syntac-
tic structures since they build on the basic relations of immediate dominance
and linear precedence.
The commas on the right-hand side of these rules do not indicate any
ordering of the subconstituents. The ordering is specified by additional
linear precedence rules, which are partly universal and partly language-
specific. For example, English, a verb-medial language, obeys the following
linear precedence rules (Van Valin and LaPolla 1997: 71):
As Van Valin and LaPolla point out, the range of possible syntactic constella-
tions specified by the above rules needs to be further constrained by the
linking of syntax to semantics, which includes constraints on the syntactic
realization of arguments depending on semantic structures in the lexicon.
In addition to the constituent structure rules in (1) and (2), a separate set of
immediate dominance and linear precedence rules is needed for specifying
the structure of the operator projection.
The idea of specifying the constituent structure and the operator structure
separately by different context-free grammars was originally proposed by
Johnson (1987), based on the observation that the ordering among the
operators is systematically correlated with their scope given by their attach-
ment site at the clausal skeleton, whereas the surface order of the operators
relative to arguments and adjuncts is much less transparent and often
requires crossing branches. The two grammars taken together then consti-
tute a projection grammar, giving rise to a constituent projection and an operator
projection. Johnson formally defines a projection grammar as a quadruple
P ¼ T; I; ðAi Þi2I , ðGi Þi2I , where T is a set of terminal symbols, I is a finite
set of ‘projection’ indices and, for every i 2 I, Ai is a subset of T and Gi is a
formal grammar with terminal symbols in Ai . A string s over T then belongs
to the language generated by P if and only if its ith projection, that is, the
concatenation of the elements of s belonging to Ai , is in the language
generated by Gi .
While it seems reasonable to distinguish between constituent structure
and operator structure, Johnson’s proposal has the problem of being purely
surface oriented. As a consequence, it does not enforce matching clausal
skeletons in the two projections. However, corresponding clausal skeletons
in both projections are taken for granted in the syntactic representations of
RRG. A further problem arises from the assumption that the operator
projection can be represented as a tree, that is, that each operator contrib-
utes only to one layer (cf. Section 20.5 for counterexamples).
Approach (ii), the second approach discussed by Van Valin and LaPolla,
postulates an inventory of elementary syntactic trees that can be combined
into more complex syntactic structures. Van Valin and LaPolla (1997: 654,
note 34) point out that Tree Adjoining Grammars (TAG, Joshi and Schabes
1997) may provide a way to formalize such tree templates and their
composition. Building on this idea, Kallmeyer et al. (2013) (see also Kall-
meyer and Osswald 2017; Osswald and Kallmeyer 2018) propose a formal-
ization of RRG as a grammar based on so-called elementary trees and TAG-
inspired tree composition operations. We will detail this approach in
Section 20.3.
A slightly different proposal has been suggested by Nolan (2004), who
argues for a formalization of RRG that systematically exploits feature-based
representations, similar in style to HPSG (Pollard and Sag 1994) and, more
recently, Sign-Based Construction Grammar (Sag 2012). Representing con-
stituent structures in this way calls for features, or attributes, by which the
subconstituents can be addressed. This can be done either by reconstructing
tree structures as feature structures based on formal features such as first
and rest, or by employing functional notions like subject, direct-object,
etc. However, in RRG, configurational syntactic notions are usually not
considered as basic but rather as derived concepts. While the representation
of the constituent projection proposed in Nolan (2004, §5.5), which builds on
the immediate dominance rules given in (1), is not fully explicit about the
attributes involved, it seems that it uses either pure list-oriented attributes
(first and rest, or even 1st, 2nd, 3rd, etc.) or attributes specific to the
categories of the subconstituents.
SENTENCE
CLAUSE
RP did RP NUC
boy
Figure 20.2 Operator marking by features
SENTENCE
CLAUSE
CORE
NUCM
blue
Figure 20.3 Periphery marking by features
In this example, the transitive verb template associated with the verbal
predicate smash contains not only the NUC node but also the CORE and
the two RP argument slots. Note that the lexical elements (marked grey in
the figure) are not part of the templates but are added in a separate step.
The availability of larger syntactic units allows one to lexicalize parts of the
grammar, up to the point of a fully Lexicalized Tree Adjoining Grammar
(LTAG), in which every elementary tree is required to have a lexical anchor.
The similarities between RRG’s tree templates and (L)TAG’s elementary
trees suggest a formalization of RRG syntax along the following lines:
A language is syntactically described by a tree rewriting grammar compris-
ing a set of tree templates, lexical elements filling the anchor nodes of these
templates, and certain operations for combining them into syntactic
CORER PRED
NUCR V DEF RP
PERIPHERY
N smashed the MP CORER
NUCM N
A vase
blue
Figure 20.4 RRG templates
SENTENCE
SENTENCE
CLAUSE CLAUSE
CLAUSE
CORE
RP CORE
RP NUC RP
RP CORER RP NUC RP
V
RP∗
CORER CORE∗R NUCR CORER V OPdef CORER
smashed
OPdef
NUCR MPperi N NUCR smashed the MPperi NUCR
the
N COREM vase N COREM N
A A
blue blue
CLAUSE RP CORE∗R
building blocks but are generated in a modular fashion from classes of tree
constraints in the meta-grammar. Since fewer ordering constraints in the
meta-grammar correspond to a larger set of elementary tree templates, it
follows, loosely speaking and without taking into account morphological
markings, that free word order is descriptively more simple in the meta-
grammar than strict word order.
(a)
(b)
Figure 20.7 Schematic sketch of simple substitution (a) and sister adjunction (b)
to the target node in the constituent tree. Their category (OP) indicates that
they are operators and their contribution is indicated within further fea-
tures attached to the OP node. Their elementary trees are rooted by a node
that merges with the target tree. An advantage of this is that one can
formulate constraints by means of feature specifications. For instance, one
can require that there is only one definiteness operator in an RP. We will see
how to use features for this purpose in more detail in Section 20.5.
Peripheral structures cannot be added by substitution since they do not
attach to leaves but to internal nodes, in general, and the same holds for the
operators; cf. Figure 20.5. The mode of composition proposed for these cases
is (sister) adjunction (see also Kallmeyer et al. 2013).4 As with substitution, we
assume that the templates available for adjunction have a root label which
coincides with the label of the target node (cf. Figure 20.5). For convenience,
the root of an adjunction tree is marked by an asterisk in the graphical
presentations. We call elementary trees with this marking adjunct trees.
A further constraint on adjunct trees is that their root node has only a
single daughter. The idea behind this is, as explained earlier, that the tree
below the root is the actual periphery element while the root node captures
more or less the PERIPHERY edge from RRG textbooks. The root label of an
adjunction tree specifies the attachment site at the phrasal skeleton. In
Figure 20.8, for example, the adverbial completely adjoins at the nucleus
while yesterday adjoins at the core.
CLAUSE
CORE
RP RP NUC RP CORE∗
MPperi RP
finished
yesterday
the cake
completely
CLAUSE
CORE
RP NUC RP MPperi
completely finished
Figure 20.8 Sister adjunction of periphery elements at different layers
Sister adjunction is defined in such a way that the target node must
not be a leaf node. Because of their single-daughter property, roots of
adjunct trees are excluded as target nodes for sister adjunction, but it is
of course possible to adjoin more than one adjunct tree to the same
node of the target tree. The adjunction operation consists in merging
the root of the adjunct tree with the target node (which amounts to
unifying their feature structures) and adding the daughter of the
adjunct tree root as a new daughter to the target node. This can take
place in any position among the already existing daughters of that
target node.5 See Figure 20.7b for a general schematic illustration of
sister adjunction.
A complication arises when an operator or periphery element targeting a
specific layer occurs between elements that are part of a different layer.
Examples are given in (4):
In (4a), the nucleus, which consists of the verb ate, is followed by an argu-
ment, which is part of the core, and after that comes an adverb that is in the
nuclear periphery. In (4b), the tense operator did is placed between an
argument (part of the core) and the verb (also part of the core). But it should
attach at the clausal level. For the moment, we will ignore this complication
and come back to it in Sections 20.5 and 20.7.
(a) (b)
SENTENCE SENTENCE
CLAUSE CLAUSE
PrCS CLAUSE
PrCS CORE CLAUSE
RP CORE CLAUSE
RP RP NUC CORE
RP NUC CORE
RP NUC RP NUC
what does John think Kim smashed what does John think Kim smashed
Figure 20.9 Two possible syntactic representations of wh-extraction from complements
(a) (b)
SENTENCE SENTENCE
CLAUSE CLAUSE
CLAUSE CLAUSE
RP RP NUC RP CORE
CLAUSE
CLAUSE
think RP NUC
CORE CLAUSE
CORE CLAUSE
smash
RP NUC
RP NUC
CLAUSE
CLAUSE think
claim
CORE CLAUSE
CORE
RP NUC
RP NUC
claim
smash
Figure 20.10 Wh-extraction via simple substitution (a) and wrapping substitution (b)
dominance edges in the final derived tree. We call such an edge a d-edge.
Under this analysis, the long-distance dependency comes about by the
insertion of material at the d-edge, that is, between the pre-core slot and
the corresponding core.6 More flexible templates and a more complicated
tree composition mechanism are needed in this case. The clause node of the
smash structure is split in two and its upper part is unified with the upper
clause node of the think structure (similar to the adjunction mechanism
introduced in the previous section), thereby keeping the pre-core slot on the
left side of the structure; the lower part of the split node is substituted at the
lower clause node of the claim structure. The mode of composition just
described is referred to as wrapping substitution (Kallmeyer et al. 2013; Oss-
wald and Kallmeyer 2018).
Wrapping substitution has the general form depicted in Figure 20.11.7 By
definition, this operation involves a tree with a d-edge. Such an edge stands
for a dominance relation in the final derived tree, that is, it specifies a
place in an elementary tree where additional nodes and edges from other
elementary trees can be inserted.8 Wrapping consists basically of splitting
the tree at the d-edge and wrapping it around a target tree. More specific-
ally, with reference to Figure 20.11, the subtree β rooted at the lower node
(labelled X) of the d-edge fills a substitution slot in the target tree α while
the upper node of the d-edge (labelled Y) merges with the root of α.
Concerning this upper node, all descendants to the left (resp. right) of
the d-edge are located to the left (resp. right) of the target tree α in the
resulting combined tree, and all nodes of γ dominating the upper node
(labelled Y) of the d-edge also dominate the merged node after the wrap-
ping. The example in Figure 20.10b is a case of wrapping with empty γ and
δR, (i.e. a wrapping where the upper part only adds nodes to the left of the
d-edge but neither to its right nor above).9
It is instructive to compare the different tree composition options of
Figure 20.10 with respect to their applicability for linking, especially with
respect to template selection. As already explained, the syntactic inven-
tory does not provide a single template for the complex syntactic struc-
ture of (5b). Rather, the structure has to be composed from argument
structure templates which in turn are selected by the chosen lexical
entries. The composition in Figure 20.10a has the disadvantages that
the template selected by think is not an argument structure template
CLAUSE
PrCS CLAUSE
RP CORE
RP NUC CLAUSE
think CORE
RP NUC CLAUSE
claim CORE
RP NUC
smash
Figure 20.12 D-edge in final derived tree for Figure 20.10b
rank k by showing that for every k-TWG, a simple CFTG of rank k can be
constructed that generates the same language. Simple CFTGs of rank k are,
in turn, equivalent to well-nested Linear Context-Free Rewriting Systems
(LCFRS) of fan-out k þ 1. Consequently, k-TWGs are in particular mildly
context-sensitive.
The notion of mildly context-sensitive languages and formalisms was
introduced in Joshi (1985) in an attempt to characterize the amount of
context-sensitivity required for natural languages, where context-sensitivity
is used in a formal language sense, in contrast to context-free languages and
grammars. Roughly, a mildly context-sensitive grammar formalism is able
to generate more than all context-free languages, can account for cross-
serial dependencies, is polynomially parsable,10 and has the constant-
growth property, which means that, if we order the sentences of a language
according to their length, the length grows in a linear way. It is interesting
that coming from RRG, a typologically motivated linguistic theory, and
developing a tree rewriting formalization of it, we end up within a class of
formalisms that seems to support Joshi’s hypothesis that natural languages
are mildly context-sensitive.
SENTENCE SENTENCE
CLAUSE CLAUSE
CORE CORE
V[ +] AP[ +] V[ +] AP[ +]
John painted the house red John painted the house red
Figure 20.13 Discontinuous complex predicates
Notice the two templates for the passive (b and c), one in which only the
undergoer occurs as an argument, and a second template which includes
the realization of the actor by a peripheral by-phrase.13 Standard RRG would
probably regard the peripheral by-phrase as a separate template that can be
adjoined to the passive core. In the context of the present framework, the
by
(d) (e) (f)
CLAUSE NUC∗N CORE
V[PRED +]
Figure 20.14 Basic transitive predication template for English with variants
V[PRED +]
clause-spine:= base-transitive :=
core-spine ∧ core-clause clause-spine ∧ pre-nuc-rp ∧ post-nuc-rp
CLAUSE CLAUSE
CORE CORE
NUC RP ≺ NUC ≺ RP
V[PRED +] V[PRED +]
Figure 20.15 Example specifications of syntactic fragments
diagrams but are to be read as tree descriptions. For instance, the specifica-
tion with the name pre-core-slot says that there are three distinct nodes n0, n1
and n2, labelled respectively by CLAUSE, PrCS and CORE, where n1 and n2 are
daughters of n0 and n1 immediately precedes n2 (expressed by ≺).
Figure 20.15 illustrates how the basic transitive template of Figure 20.14
can be defined by a piecewise combination of such specifications. The pre-
core slot template of Figure 20.14 can likewise be defined by conjoining
the specifications clause-spine, pre-core-slot, and pre-nuc-rp. In this way, common
components of elementary templates are made accessible in the meta-
grammar.14 Furthermore, syntactic tree descriptions can be linked to
descriptions of semantic representations (either predicate-logical formulas
or frames), and the meta-grammar can include a constraint-based formula-
tion of RRG’s linking algorithm (cf. Kallmeyer et al. 2016). The syntax–
semantics interface will be briefly discussed in Section 20.8.
Layer Operators
away from the nucleus the operator occurs on the surface. The mapping
from operators to levels of the layered structure explains (i) the scope
behaviour of operators, since structurally higher operators take scope over
lower ones, and (ii) surface order constraints for operators; higher operators
are further away from the nucleus of the structure.
The problem is that the constituent and the operator structure are not
completely parallel; one can have structures where an operator belonging to
a specific layer is, on the surface level, surrounded by elements belonging to
a lower layer in the constituent structure. Examples of a clause-level tense
operator occurring within the core are given in (7) and (8) (taken from Van
Valin, 2005: 10) for English and Turkish, respectively.
In (8), the clause-level status and tense operators occur between the verb and
the pronominal affix, which is part of the core.
Even though the constituent structure and the operator structure are not
fully aligned, they depend on each other. Their hierarchical order is the
same and the existence of a layer in the operator projection requires that
this layer also exists in the constituent structure. For instance, one can only
have clause-level operators if a clause node exists in the constituent
structure.
While the ordering among the operators is thus systematically correlated
with the scope given by their attachment site at the clausal skeleton, the
surface order of the operators relative to arguments and adjuncts is much
less transparent and would require crossing branches if everything were
captured in a single tree with operators attaching where they take scope. For
this reason, RRG usually represents the constituent structure and the oper-
ator structure as different projections of the clause. The syntactic represen-
tation for (7) on the left side of Figure 20.16 illustrates this idea: the upper
part gives the constituent projection while the lower part gives the operator
projection. If we integrate the operator projection into the clausal skeleton
of the constituent structure, we obtain the tree with crossing branches
shown on the right of Figure 20.16, due to the above-mentioned mismatches
between the two projections.15
Besides crossing branches, additional complications arise from the
following two facts: first, a functional element can contribute more than
one operator and, second, an operator can be distributed over more than
one element in the sentence. An example is (9) where the functional element
will contributes both tense and illocutionary force (IF), while aspect (ASP) is
jointly contributed by be and -ing.
CLAUSE CLAUSE
CORE
CORE
RP NUC RP
V NUC
RP RP
Mary enter-ed the room V TNS
V
Mary enter -ed the room
NUC
CORE
CLAUSE TNS
Figure 20.16 Constituent structure and operator projection for (7)
CLAUSE
CORE
CLAUSE
RP NUC
V
will they be leav -ing IF TNS CORE
V RP NUC
RRG assumes the structure on the left of Figure 20.17 (adapted from Van
Valin 2005: 14). If the operator projection is integrated into the constituent
structure, we obtain the graph on the right of Figure 20.17, which is not
even a tree but only a directed acyclic graph since the node labelled will has
two incoming edges. Note, however, that as long as nodes with more than
one incoming edge arise only from contributing several operator categories
to the same level of the layered structure, the graph can easily be turned
into a tree by putting all these categories into a single node with a label (or,
rather, a feature structure) that captures the fact that the element is an
operator that contributes several categories to the same layer. The node
dominating will in Figure 20.17, for instance, could have a category OP with
features TNS ¼ fut and IF ¼ int.
Cases with a single element contributing two operators that take scope at
different layers exist as well. Examples are finite modal verbs as in (10) that
provide modality at the core level and tense at the clause.
CLAUSE
CORE
CLAUSE
RP NUC
V CORE
John might win
V RP NUC
keeping track of the operator projection within the feature structures. The
classification of an element as an operator and the contribution of that
element to the various layers is captured within the category of its node.
More specifically, an operator element has the category label OP and a
feature structure that specifies its contribution to the CLAUSE (feature cl),
the CORE (feature co) and the NUC (feature nuc) layer. The tree structures
corresponding to the sentences (9) and (10) are shown in Figure 20.19. Note
that the operator scope information is now entirely captured within the
feature structure at the OP node and does not depend on the category of the
node to which the OP attaches.
In order to obtain trees of the type given in Figure 20.19, we use sister
adjunction for adding operators. This leads to spurious ambiguities since
operators could in principle adjoin to any of the three layers that they
attach to, provided this does not yield crossing branches. In the first tree
in Figure 20.19 for instance, we could also attach the clausal operator will
lower, at the CORE node, and be could attach higher, also at the CORE
node. In the following we assume that among the different combinations
of attachment sites that are possible, the preferred one is the one where
operators are placed as close as possible to the layer that corresponds to
their scope (if it scopes at several layers, the lowest is considered
relevant).
Aside from this spurious ambiguity issue, if we adjoin operators in an
unrestricted way, it leads to undesired overgeneration since we can leave
CLAUSE
IF int
OP CL
TENSE fut
CORE
will RP NUC
be leav -ing
CLAUSE
CORE
CL [TENSE pres]
RP OP CO [MOD epistemic]
NUC
John might V
win
Figure 20.19 Encoding the operator structure in node features for (9) and (10)
out operators that are obligatory, adjoin several operators of the same type
even in cases where this is ungrammatical, and we can generate ungram-
matical linear orders of operators, for instance the ones in (11).
CLAUSE
does
Figure 20.20 Obligatory adjunction of a tense operator
2 [: : : ] • • 3 [: : : ]
• 1 [: : : ] 1 [: : : ] • 2 [: : : ] • • 3 [: : : ]
of the PrCS signals that to the right of this node, a tense operator has to be
present. If no tense operator adjoins, the absence and the requirement of the
presence will eventually unify (we will explain later why) and this will lead
to a unification failure. If, however, a tense operator adjoins as in
Figure 20.20, the requirement is satisfied. The features to the left and to
the right of the tense operator tell us that to its right, no other tense
operator exists while to its left one can pass the information that there is
now a tense operator. (OPtns is short for category OP with feature structure
[cl [tense . . .]].)
Let us explain now exactly how edge features work (see Kallmeyer and
Osswald 2017), in particular how they percolate through the tree (cf.
Figure 20.21). As mentioned, nodes can have special left and right feature
structures. In the final derived tree, the left feature structure of a node ν
unifies with the right feature structure of its immediate sister to the left.
Furthermore, the left feature structure of a node ν that does not have a sister
to the left unifies with the left feature structure of the mother of ν, provided
this mother is not the root node of an elementary tree or the lower node of a
d-edge. Similarly, the right feature structure of a node ν that does not have a
sister to the right unifies with the right feature structure of the mother of ν,
again provided this node is not the root node of an elementary tree or the
lower node of a d-edge. In our example, once we have performed the
adjunction, we obtain the tree at the top of Figure 20.22, and after the final
edge feature unifications, the result is the tree at the bottom. Such edge
features can be used to require certain adjunctions and, as is the case in this
example, to require them exactly once.
The adjunction of the operators needs to be controlled in the following two
respects: (i) Adjoining an operator is obligatory if the information conveyed by
the operator is required for a sentence to be complete. (ii) The scope-related
ordering of the operators must be respected. In our approach, these constraints
are both implemented with the help of edge feature structures; we have just
seen an example of type (i). We will now introduce features that guarantee (ii)
as well. Concretely, we want to achieve that, from the nuclear predicate
outwards, one encounters first the NUC operators, then the CORE operators
and then the CLAUSE operators. To this end, we assume an edge feature ops
(for operator structure) that specifies which operator projection layer(s) have
say
say
been reached so far. Its value is a feature structure with features cl_l, co_l
and nuc_l for the three layers, each with possible values þ or . These
features are used in such a way as to guarantee that nuclear, core and
clausal operators have to appear in this order when moving outwards in
the sentence, starting from the nuclear predicate.
Figure 20.23 illustrates this mechanism by applying it to the operator
structure of the sentence in (12).
For instance, a core operator such as not that adjoins to the left of the
predicate has a requirement to the right that the level cl_l have a value
(i.e. has not been reached yet). To the left, it just gives the information
co_l þ, which has the effect of disallowing nuclear operators to appear
here. The operator might, which contributes not only tense at the clausal
level but also modality at the core level, comes with the same require-
ments to its right. But, in contrast to not, it states to the left of the edge
that now cl_l has value þ (as well as co_l). An inverse order (John not
might win) is therefore excluded. To the left of might, there cannot be any
further core (or nuclear) operators. The derived tree for sentence (10) is
shown in Figure 20.24.
Recall that in addition to the ops feature, nodes of category OP have
features nuc, co and cl that indicate the contribution of the operator.
For example, since might contributes tense at the clause level and epistemic
modality at the core level, its OP feature structure is [cl[tense pres], co[mod
epistemic]] (cf. Figure 20.19). (We left this out of Figures 20.23 and 20.24 for
CLAUSE
TNS + CORE
TNS 3 TNS 3
OPS 4
RP OPS 4
NUC
⎡ ⎤ ⎡ ⎤
TNS −
⎡ ⎤ TNS −
⎡ ⎤
⎢ NUC_L − ⎥ ⎢ NUC_L − ⎥
John ⎢ ⎥ V ⎢ ⎥
⎣OPS ⎣CO_L − ⎦⎦ ⎣OPS ⎣CO_L − ⎦⎦
CORE ∗ CL_L − CL_L −
⎡ ⎤
TNS + win
⎣ TNS −
CL_L + ⎦ OPmod,tns
OPS OPS CL_L −
CO_L +
might
CORE∗
TNS − TNS −
OPS [CO_L +]
OPneg OPS CL_L −
not
Figure 20.23 Keeping track of the operator projection in edge features
the sake of readability.) Since the features of the OP nodes specify which
operator projection layer(s) the operator belongs to, we can deterministic-
ally map a derived tree to the standard RRG structure in which the constitu-
ent structure and the operator projection are separated.
CLAUSE
TNS + CORE
⎡ ⎤
TNS +
TNS 3 TNS 3 ⎣ TNS − TNS − TNS −
RP CL_L + ⎦ OPmod,tns OPneg NUC
OPS 4 OPS 4 OPS OPS CL_L − OPS [CO_L +] OPS CL_L −
CO_L +
⎡ ⎤ ⎡ ⎤
TNS −
⎡ ⎤ TNS −
⎡ ⎤
⎢ NUC_L − ⎥ ⎢ NUC_L − ⎥
John might not ⎢ ⎥ ⎢ ⎥
⎣OPS ⎣CO_L − ⎦⎦ V ⎣OPS ⎣CO_L − ⎦⎦
CL_L − CL_L −
CLAUSE
⎡ ⎤
win
⎡ ⎤ TNS −
TNS + ⎡ ⎤
⎢ CL_L − ⎥
⎣ CL_L + ⎦ CORE ⎢ ⎥
OPS ⎣OPS ⎣CO_L − ⎦⎦
CO_L +
NUC_L −
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ TNS − TNS − TNS −
TNS + TNS + TNS + TNS − TNS − ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
⎢ NUC_L − ⎥ ⎢ NUC_L − ⎥ ⎢ NUC_L − ⎥
⎣ CL_L + ⎦ RP ⎣ CL_L + ⎦ ⎣ CL_L + ⎦ OPmod,tns ⎣ CL_L − ⎦ ⎣ CL_L − ⎦ OPneg ⎢
⎣OPS ⎣CO_L − ⎦⎦
⎥ ⎢ ⎥
⎣OPS ⎣CO_L − ⎦⎦ NUC
⎢
⎣OPS ⎣CO_L − ⎦⎦
⎥
OPS OPS OPS OPS OPS
CO_L + CO_L + CO_L + CO_L + CO_L +
CL_L − CL_L − CL_L −
⎡ ⎤ ⎡ ⎤
TNS −
⎡ ⎤ TNS −
⎡ ⎤
⎢ NUC_L − ⎥⎥ ⎢ NUC_L − ⎥
John might not ⎢ ⎢ ⎥
⎣OPS ⎣CO_L − ⎦⎦ V ⎣OPS ⎣CO_L − ⎦⎦
CL_L − CL_L −
win
Figure 20.24 Derived tree for (10) before (top) and after (bottom) final edge feature unification
Formalization of RRG Syntax 767
X Y
… …
X X … X X X … X
Figure 20.25 RRG template schemas for non-subordinate nexus types
sister adjunction
CLAUSE
Figure 20.25 (cf. Van Valin and LaPolla 1997: 507 and Van Valin 2005: 224).
However, these structures cannot serve as elementary trees in the sense
introduced in Section 20.3 since there are infinitely many of them; that
is, they have to be derived by operations in the syntax.
Among the tree composition operations described in Section 20.3, both
sister adjunction and wrapping substitution allow the derivation of cosu-
bordinate structures, as sketched in Figure 20.26 for the case of core
cosubordination. The following two examples of multi-verb constructions
show possible use cases for the two kinds of derivations. Both examples
describe transitive motion scenarios. The example in (13), taken from
Ullrich (2011), illustrates a common pattern for simultaneous event con-
structions in Lakhota (Siouan) (cf. Ullrich 2018; Osswald and Van Valin
2022).
CLAUSE
CORE[COSUB +]∗
COSUB + CORE[COSUB +]
COSUB + CORE
COSUB − CORE
NUC
NUC
V
V
(a)
CLAUSE
CORE
CORE CORE
RP RP NUC RP NUC
(b)
CLAUSE
RP RP NUC RP NUC
korogashi-te ire-
Figure 20.28 Core cosubordination analysis of (14) derived by wrapping substitution
SENTENCE
CLAUSE
CORE
CORE
SENTENCE CORE CORE
CLAUSE
RP NUC LM NUC RP LM NUC RP
CORE
CORE John tried to persuade Kim to clean the floor
CORE CORE
Structures like the ones in Figure 20.29 show a mismatch between syntax
and semantics in the following sense: semantically, the infinitival
(a)
CLAUSE CLAUSE CLAUSE
(b)
CLAUSE CLAUSE
CLAUSE
CORE CORE CORE
CORE CORE CORE CORE CORE CORE
Figure 20.30 Composition of templates by wrapping substitution for (15a) and (15b)
CORE CORE PrCS CORE CORE CORE PrCS CORE CORE CORE
(16) a. Whom did Mary expect John to ask to clean the floor?
b. What did Mary expect John to ask Kim to clean?
(a)
CLAUSE CLAUSE
RP NUC RP NUC RP
Kim smashed
Figure 20.32 Subordination via simple and wrapping substitution for (17a) and (17b), respectively
CLAUSE
CORE[MOD deont]
V ip V meli yiz
gid gör
Figure 20.33 Derived tree for (18)
Let us assume that the first core is added to the second core by sister
adjunction, similar to the analysis of the Lakhota construction in
Figure 20.27. The modal operator in (18) adjoins to the second embedded
core node and carries a feature indicating that it is a core operator. The
result should be the derived structure in Figure 20.33 with the mod feature
shared between all the three CORE nodes involved (cf. Van Valin, 2005:
204).16 In a cosubordination, an operator embedded in one part of the
complex structure generally takes scope over the larger category. Accord-
ingly, in all elementary trees for cosubordination configurations, the rele-
vant features (here mod) are shared between the lower and the higher
category in question (here the two CORE nodes). This is taken to be a general
property of cosubordination structures. Corresponding to this, we assume
that when mapping our derived structure to the standard RRG structure, the
operator targets the highest corresponding node, as long as there is no
higher operator level and no substitution node in between. In the case of
Figure 20.33, this is the core of the entire sentence.
Figure 20.34 gives the derivation of the tree in Figure 20.33, attaching the
tree for gid ip by sister adjunction. Note that this implies that there is a
special tree for cores such as gör yiz in this example, providing two CORE
nodes and requiring an adjunction at the higher one in order to be com-
pleted to a cosubordination structure. This requirement is again expressed
by an edge feature cosub. We can adjoin several cores such as gid ip but we
have to adjoin at least one in order to switch the feature from to þ.
Besides the edge feature, we also assume (as before, see Figure 20.27) a
boolean node feature cosub that expresses whether a node roots a cosubor-
dination structure or not (see also the CORE nodes in Figure 20.34). This
enforces an adjunction of the gid tree at the cosubordination CORE node and
not at the lower CORE node.
CLAUSE
COSUB + COSUB +
CORE ∗ COSUB + CORE
MOD 1 MOD 2
COSUB − COSUB −
COSUB + CORE COSUB − CORE
MOD 1 MOD 2
V ip V yiz
meli
Figure 20.34 Derivation for (18)
(19) a. [[Kim mustMOD go]CORE [to try]CORE [to wash the car]CORE]CORE
b. [[Kim mustMOD ask Pat]CORE [to wash the car]CORE]CLAUSE
Concerning the analysis of (19a), we propose that the different cores are
combined with each other via wrapping substitution and not via sister
adjunction. The difference, compared to (18) is that the verbs go and try
both syntactically select for an infinitival complement clause (i.e. for a
CORE argument), which should be expressed via a corresponding substi-
tution node. The derivation is shown in Figure 20.35. The node features
contributed by CORE operators (here [mod deont]) are shared between the
different CORE nodes that are part of the cosubordination construction.
An edge feature for enforcing the adjunction of further CORE sisters in a
cosubordination is not necessary since the substitution nodes guarantee
that CORE arguments have to be added.
The shared operator scope in (19a) is a standard criterion for distinguish-
ing cosubordinate from coordinate constructions. Another diagnostic is the
COSUB +
CORE
MOD 1
COSUB − COSUB −
CORE CORE
MOD 1 MOD 1
CLAUSE
RP NUC
COSUB +
CORE
MOD 2 COSUB +
Kim V CORE
MOD 3
go COSUB − COSUB −
CORE CORE
MOD 2 MOD 2 COSUB −
COSUB − ∗ CORE
CORE MOD 3
MOD deont
to try
OP[CO [MOD deont]] to wash the car
must
Figure 20.35 Derivation for (19a)
(20) a. Kim must ask Pat now to wash the car tomorrow
b. #Kim must go now to try to wash the car tomorrow
c. #Kim must go to try now to wash the car tomorrow
The substitution of the clausal argument into the tree anchored by told is
shown in Figure 20.36. The operator will in the embedded clause contributes
tense at the clausal layer. The corresponding feature [cl [tense fut]] is shared
along the clausal skeleton of the embedded sentence, via the ½op 2 features.
But it is not transported into the embedding clause, which can have a
different tense feature. The edge feature percolation also stops at the substi-
tution node, which means that the tense in the embedded clause cannot
satisfy a tense requirement in the embedding clause.
⎡ ⎤
TNS +
⎣ CL_L + ⎦ OP 1 [CL [TENSE fut]] TNS −
OPS
CO_L +
will
Figure 20.36 Derivation for (21)
Adjuncts in RRG are part of the periphery, which can be seen as a structure
similar to the operator projection since each adjunct targets a specific layer
and, starting from the nucleus and moving outside, the nuclear adjuncts
have to precede the core adjuncts, which in turn have to precede the clausal
adjuncts. Kallmeyer and Osswald (2017) briefly mention that this can be
modelled in a way similar to the treatment of operators explained above. In
the following, we will illustrate with an example how to capture the
ordering constraints for periphery elements via edge features in a way
analogous to the corresponding word-order constraints for operators. We
will restrict ourselves to adverbs but other modifiers would be treated in a
similar way.
According to Van Valin (2005: 41), adverbs may ‘modify all three layers
of the clause; aspectual adverbs like completely and continuously modify the
nucleus, pace adverbs like quickly and manner adverbs like carefully
modify the core, and epistemic adverbs like probably and evidential
adverbs like evidently modify the clause’. Depending on their position,
manner adverbs can actually be clausal or core modifiers, see (22), from
Van Valin (2005).
The core modifier reading signifies that Ruth hid the cash in a clever way
while the clausal modification has the meaning that it was clever of Ruth to
hide the cash. (22a) is ambiguous between the two readings while (22b)
seems to have only the ‘in a clever way’ reading and (22c) only the wide-
scope reading (‘it was clever to hide the cash’).
As to linear precedence, the main difference between modifiers and oper-
ators is that modifiers can often be placed to the left or to the right of the
nucleus (see (22)). Moreover, modifiers differ from operators regarding con-
straints on adjunction: they are rarely obligatory and there can be multiple
modifiers targeting the same layer, as in (23). If multiple modification is not
possible, this is due to semantic or pragmatic ill-formedness (e.g. conflicting
aspectual information).
(23) a. Ruth carefully hid the cash slowly. (several core modifiers)
b. Evidently, Ruth possibly hid the cash. (several clause modifiers)
CLAUSE
CORE
RP OP OP NUC RP PP
immersing
Figure 20.37 Syntactic tree for (24)
CLAUSE
CORE
immersing
CORE∗ CORE∗
evidently slowly
(25) a. Leslie has evidently been slowly immersing herself in the new language.
b. *Leslie has slowly been evidently immersing herself in the new language.
The main focus of the present chapter is on formalizing the syntactic side of
RRG. This section shows briefly how the described tree rewriting approach
can be extended to combining syntactic and semantic composition along
the lines of Kallmeyer and Osswald (2013) and Kallmeyer et al. (2016). In this
approach, the linking between syntactic and semantic components is
CLAUSE
CORE[IND e]
smashed
e = [do′(x, Ø)] CAUSE [BECOME smashed′(y)], x = Kim, y = vase
⎡ ⎤
causation
⎢ ⎡ ⎤ ⎥
⎢ activity ⎥
⎢ ⎣ ⎦ ⎥
⎢ person ⎥
⎢ x ⎥
⎢ ‘Kim’ ⎥
e⎢
⎢
⎡ ⎤⎥
⎥
⎢ change_of_state ⎥
⎢ ⎢ ⎥⎥
⎢ ⎢ y vase ⎥⎥
⎢ ⎢ ⎥⎥
⎣ ⎣ smashed_state ⎦⎦
y
CLAUSE
CORE[IND e]
⎡ ⎤
RP[IND x] NUC[IND e] RP[IND y] causation
⎢ ⎥
⎢ activity ⎥
⎢ ⎥
V[IND e] ⎢ x ⎥
⎢ ⎡ ⎤⎥
e⎢
⎢ change_of_state ⎥
⎥
⎢ ⎢ ⎥⎥
⎢ ⎢ y ⎥⎥
V[IND e ] ⎢ ⎢ ⎥⎥
⎣ ⎣ smashed_state ⎦⎦
y
smashed
CLAUSE ⎡ ⎤
causation
⎢ ⎥
CORE[IND e] ⎢ activity ⎥
⎢ ⎥
⎢ x ⎥
⎢ ⎡ ⎤⎥
⎢
e⎢ change_of_state ⎥
⎥
RP[IND x] NUC[IND e] RP[IND y] ⎢ ⎢ ⎥⎥
⎢ ⎢ y ⎥⎥
⎢ ⎢ ⎥⎥
⎣ ⎣ smashed_state ⎦⎦
V[IND e]
y
smashed
RP[IND x ] RP[IND y ]
20.9 Conclusion
References
Osswald, Rainer and Robert D. Van Valin, Jr. 2014. FrameNet, frame struc-
ture, and the syntax–semantics interface. In Thomas Gamerschlag, Doris
Gerland, Rainer Osswald and Wiebke Petersen (eds.), Frames and Concept
Types, 125–156. Dordrecht: Springer.
Osswald, Rainer and Robert D. Van Valin, Jr. 2022. The description of transi-
tive directed motion in Lakhota (Siouan). In Laure Sarda and Benjamin
Fagard (eds.), Neglected Aspects of Motion Events Description: Deixis, Asymmetries,
Constructions, 209–233. Amsterdam: John Benjamins.
Pollard, Carl J. and Ivan A. Sag. 1994. Head-Driven Phrase Structure Grammar.
Chicago, IL: University of Chicago Press.
Rambow, Owen. 1994. Formal and Computational Aspects of Natural Language
Syntax. PhD thesis, University of Pennsylvania.
Rambow, Owen, K. Vijay-Shanker and David Weir. 1995. D-tree grammars. In
Proceedings of the 33rd Annual Meeting of the Association for Computational
Linguistics, 151–158. Association for Computational Linguistics.
Rambow, Owen, K. Vijay-Shanker and David Weir. 2001. D-tree substitution
grammars. Computational Linguistics 27(1): 87–121.
Sag, Ivan. 2012. Sign-Based Construction Grammar: An informal synopsis. In
Hans Boas and Ivan Sag (eds.), Sign-Based Construction Grammar, 61–188.
Stanford, CA: CSLI Publications.
Sag, Ivan A. and Thomas Wasow. 1999. Syntactic Theory: A Formal Introduction.
Stanford, CA: CSLI Publications.
Ullrich, Jan (ed.) 2011. New Lakota Dictionary (2nd ed.). Bloomington, IN:
Lakota Language Consortium.
Ullrich, Jan. 2018. Modification, Secondary Predication and Multi-Verb Construc-
tions in Lakota. PhD thesis, Heinrich Heine University Düsseldorf.
Van Valin, Robert D., Jr. 2005. Exploring the Syntax–Semantics Interface. Cam-
bridge: Cambridge University Press.
Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning
and Function. Cambridge: Cambridge University Press.
Notes
21.1 Introduction
particular for their needs. The RRG model is rapidly evolving to new levels of
descriptive, explanatory and computational adequacy. By computational
adequacy we mean the degree to which a theoretical model can be formally
specified to enable computational modelling as a benchmark for that
theory. As will become clear, we find that the RRG model is most suitable
as a basis for computational implementation in software and can serve as a
rich testing ground for the linguistic theory. Computational models of RRG,
by intention linguistically motivated, deal with problems at the interfaces
between concept, semantics, lexicon, syntax and morphology, including
language processing and grammar. The challenge is to determine how best
to design and implement the model and its principled interfaces
in software.
What makes RRG suitable for implementation in software is that it is a
model of grammar that posits a direct mapping between the semantic
representation of a sentence and its syntactic representation (Van Valin
2005, 2008). RRG is a monostratal theory, positing only one level of syntac-
tic representation. The actual form of the sentence and its linking algo-
rithm can work in both directions from syntactic representation to
semantic representation and vice versa. In RRG, semantic decomposition
of predicates and their semantic argument structures are represented as
logical structures. The lexicon in RRG takes the position that lexical entries
for verbs should contain unique information only, with as much infor-
mation as possible derived from general lexical rules. RRG makes use of a
set of thematic roles organized into a hierarchy in which the highest-
ranking roles are Actor (for the most active participant) and Undergoer
(for most affected).
Amongst the design challenges is the requirement for a digital model
of the lexicon as a persistent store, either as a database, persistent XML
tree structure or serializable object. The architecture of the lexicon is
then an empirical problem for the software designer. It is clear that while
the RRG lexicon is traditionally considered with respect to verbal logical
structures, the software typically needs the lexicon to also include nouns,
adjectives, prepositions, and so on, in some meaningful way. To deal with
many of the problems found in modelling language within the model,
constructions also need to be stored as occurrence patterns. Additionally,
lexemes need to be meaningfully stored within the lexicon. As grammat-
ical morphemes are very much language-specific, a morpheme store is
also needed within the software. All the software and computing applica-
tions mentioned in the case studies make extensive use of the bidirec-
tional linking system from morphosyntax to semantics and semantics to
morphosyntax. The implementation may follow construction-oriented
processing according to some language-specific set of rules translated
into computer code.
The reasons why many computer scientists and computational linguists
use RRG are precisely because of the desire to deploy a functional linguistic
model that has a strong theory of the lexicon and a role for constructions.
Important also is the idea of projecting from the lexicon in a bidirectional
linking system, the layered structure of the word (LSW), the layered struc-
ture of the noun phrase (LSNP), and the layered structure of the clause with
its sub-model of nexus–juncture relations. The approach of RRG to argument
realization is highly principled in a typologically coherent way and this adds
considerable value to all language modelling efforts in software. Addition-
ally, the interfaces within RRG that reach from lexicon through semantics
via morphosyntax into speech act, information structure, pragmatics and
discourse are considered to be highly valuable.
The RRG linguistic model strives for high levels of descriptive, explana-
tory, predictive and typological adequacy (Butler 2009). Through the
deployment of the RRG model in computational applications it is shown
to meet computational adequacy. In many respects with regard to compu-
tational implementation, RRG can be considered as a ‘functional genera-
tive’ model with parse and generate ‘directions’ within the bidirectional
linking system. It does not, however, express itself as treating the
Chomsky-typed languages as formal constructs. It is more typologically
sensitive than that. While RRG can be modelled within procedurally
described or declarative approaches, it is always as a semantically motiv-
ated and lexically projected model of language. It may make use of
attribute-value matrices (AVMs) and constructions, depending on the
object of study within the computational model.
The chapter has the following organIsation. In Section 21.2, we charac-
terize the start of computational work on RRG, concentrating on the
syntactic side of the linking system. We look at the use of general-
purpose programming languages to create parsers for English, German,
Dyirbal and Biblical Hebrew. In Section 21.3, we give a description of an
RRG-based machine translation engine for Modern Standard Arabic that
parses a sentence in the native Arabic orthography and translates it into
grammatical English. The software employs an interlingua bridge archi-
tecture. Section 21.4 outlines current work on intelligent conversational
agents and humanoid avatars, where RRG is used as the language engine.
In Section 21.5, the FunGramKB project and related toolsets oriented
towards the RRG model are discussed. In Section 21.6, the contribution
of RRG to linguistic modelling in natural language processing (NLP)
applications is discussed.
Computational work using RRG has been ongoing since 2004 in a variety of
ways (Nolan 2004). Initial concerns with computational treatments of RRG
were with exploring the possibilities of expressing the theory in a compu-
tationally tractable way as a feature-based unification grammar. This
The parser has been tested by Guest on example sentences from Dyirbal,
English and Dutch. This proof-of-concept parsing method for RRG, using
templates, can handle both fixed word order and variable word order with-
out any difficulty.
Diedrichsen (2014: 105–142) describes an implementation of a sentence
parser for German that implements the linking system of RRG from syntax
to semantics (see also Diedrichsen 2008; Van Valin and Diedrichsen 2006),
and an implementation of the RRG lexicon in a data structure. This
RRG parser for German employs the concept of a construction as a
SYNTAX:
Bracket Structure: Bracket is ALWAYS opened by the finite verb. The right bracket is filled according to
the following rules:
1. If VFIN ¼AUX
{PART.OF.VERBAL.EXPRESSION ¼ full verb (PSTP)};
2. If VFIN ¼full verb
{PART.OF.VERBAL.EXPRESSION ¼ separable prefix};
3. If VFIN ¼light verb
{PART.OF.VERBAL.EXPRESSION ¼ rest of collocation};
Order of arguments within brackets follows topic comment structure and principle of end
weight. The following rules apply:
1. General constraints: pronoun > other, RP > PP
2. Case-based argument ordering constraint: NOM > DAT > ACC (default)
3. If ACC ¼ pronoun, then ACC > DAT (default)
MORPHOLOGY:
AUX: May be any auxiliary or modal verb
VFIN may be any verb.
SEMANTICS: [þtelic], where telicity may be only invoked for conversational purposes: adds intensity,
expression
PRAGMATICS:
Turn taking: Signal for dimension of Turn Constructional Unit (TCU);
right brace marks end of TCU
Illocutionary force: not specified
Focus structure inside braces: TOP>FOC
of RRG in software. In the analysis of Arabic within the software, the lexical
and grammatical properties of the Arabic words are extracted automatically.
From the parse, UniArab then creates a computer-based representation for the
logical structure of the Arabic sentence(s).
In this application, RRG has been used to motivate the computational
implementation of the lexicon as an XML data structure and to implement
in software the RRG bidirectional linking system to build the parse and
generate functions between the syntax–semantic interfaces. Through seven
input phases, including the morphological and syntactic unpacking, Uni-
Arab extracts the logical structure of an Arabic sentence. Using the XML-
based metadata representing the RRG logical structure, UniArab then accur-
ately generates an equivalent grammatical sentence in the target language
of English through four output phases. Nolan and Salem (2011) report on the
development of this machine translation system as a rule-based semantic-
ally oriented interlingua bridge framework for machine translation of
Arabic language processing using the RRG linguistic model. The UniArab
system is able to analyse intransitive, transitive and ditransitive Arabic
sentences in native Arabic (right to left) orthography and extract their
logical structure. Through a detailed study of the Arabic language, an
analyser was developed that can successfully process many of the unique
features and challenges present in Arabic. This logical structure is then used
in the generation phase, where the sentence is translated into another
language, in this case, English. The Arabic language is written from right
to left and it has complex, language-specific grammar rules and a relatively
free word order. These distinguishing features pose a major challenge in
processing Arabic text for linguistic analysis. The UniArab framework dem-
onstrates that RRG is a foundation for building multi-language machine
translation systems.
It is common to categorize the three different MT methodologies as direct,
transfer and interlingua. These levels of analysis are illustrated by what has
become known as the Vauquois triangle (Vauquois 1968) (see Figure 21.1). The
methodologies differ (i) in the linguistic depth of analysis of the source
language and (ii) in the degree to which they attempt to reach a language-
independent representation of linguistic meaning of the source and target
languages. An interlingua is designed to be a language-independent represen-
tation from which translations can be generated to different target languages.
According to Dorr (1992: 46), in order to adopt an interlingua approach to
machine translation, one must construct a language-independent represen-
tation that lends itself readily to the specification of a systematic mapping
that operates uniformly across all languages. Dorr describes such a repre-
sentation based on the lexical conceptual structure of Jackendoff (1990),
which abstracts away from syntax sufficiently to enable language-
independent encoding, while retaining enough structure to be sensitive to
the requirements for language translation. Machine translation using an
interlingua approach typically involves semantic analysis of the source
Figure 21.1 The Vauquois triangle (based on Dorr et al. 2006: 384)
language (Dorr et al. 2006: 1). At the interlingua level, a single underlying
meta-representation represents the underlying semantics of the source lan-
guage SL1 text., which will be used to generate a translation into a target
language TL1. In principle, an interlingua representation of a sentence
contains sufficient information to allow generation in any language; the
more target languages there are, the more valuable an interlingua becomes.
When there are many source languages (SLn) and many target languages
(TLn) for translation, an interlingua bridge is deployed. This has an advantage
in facilitating the scaling up to multiple languages. To add a language to an
interlingua bridge, one needs to add, for that language, a lexicon, a parse
function into a meta-representation, and a generate function from the meta-
representation. To translate from one source into N target languages, one
needs (1 þ N) steps using an interlingua. However, to translate pairwise among
all the languages, one needs only 2N steps using an interlingua compared to
about N2 with transfer – this is a significant reduction in development, design
and processing cost. In addition, it is not necessary to consider the properties
of any other language during the analysis of the source language or gener-
ation of the target language; each analyser and generator can be built inde-
pendently by a monolingual development team. Each system developer only
needs to be familiar with their own language and the interlingua. An inter-
lingua MT system will include one lexicon for each language. An advantage of
the interlingua approach is that interlingua representations have the poten-
tial to be used by other multilingual NLP applications. More importantly,
using the logical structures of RRG as the interlingua meta-representation
Figure 21.2 MT–RRG-based interlingua bridge approach (from Nolan and Salem 2011: 315)
Figure 21.3 The RRG interlingua for UniArab (from Nolan and Salem 2011: 315)
Figure 21.4 The conceptual architecture of the UniArab system (from Nolan and Salem 2011: 321)
lexicon, which holds all attributes of a word. UniArab was developed with
the lexicon encoded in XML, and also stores all data in XML format. The
system can understand the part of speech of a word, agreement features,
number, gender and the word type. The syntactic parse unpacks the agree-
ment features between elements of the Arabic sentence into a semantic
representation (the logical structure) of the state of affairs of the sentence.
The structure of the UniArab system and its operational phases are
described next.
Phase 1. Input of Arabic language sentence: The input to the system consists
of one or more sentences in Arabic. UniArab can accept as input
one or multiple sentences in a single input.
Phase 2. Sentence Tokenizer: Tokenization is the process of demarcating
and classifying sections of a string of input characters. In this
phase, the system splits the text, which may consist of paragraphs,
into sentence tokens. The resulting tokens are then passed to the
word tokenizer phase.
Phase 3. Word Tokenizer: In this phase, sentences are split into word tokens.
The Arabic sentence is read from right to left.
Phase 4. Lexicon XML Data-source: The lexicon is constructed as a set of XML
documents for each component category of Arabic. The lexicon
contains the Arabic lexemes in this XML data source ordered by
each lexical category.
Phase 5. Morphology Parser: This works directly with both the lexicon and
tokenizer. A connection is made to the lexicon data-source of
Phase 4. To understand the morphology of each word, the software
tokenizes each sentence and determines the word relationships.
This phase of the system holds all attributes specific to each word
of the source sentence.
Phase 6. Syntactic Parser: Determines the structure of the Arabic sentence.
At this point, the types and attributes of all the words in the
sentence are known.
Phase 7. Syntactic linking (RRG): The software develops the link from syntax
to semantics out of the phrasal structure created in Phase 6 to
create a logical structure for later use in generation into the target
language, and also to act as the link in the opposite direction from
semantics to syntax.
Phase 8. Logical Structure: The creation of logical structure is the most
crucial phase. An accurate representation of the logical structure
of an Arabic sentence is the primary strength of UniArab.
input, the universal logical structure of the input sentence(s), and produces,
as output, the grammatically correct morphosyntax of the target language.
In the UniArab system, Phases 9 to 12 are for generation of the target
language, which, in this case, is English.
Phase 9. Semantics to Syntax: Once the software has an input sentence and
has produced a structured syntactic representation of it, the
grammar can map this structure from a semantic representation.
In this phase, the system uses the linking algorithm of RRG to
determine Actor and Undergoer assignments, assign the core
arguments and assign the predicate in the nucleus. It determines
the grammatical ‘subject’ by analysing the agreement marking on
the verb and the various nominals. The system hinges on the use
of logical structures at the metalevel.
Phase 10. Syntax Generation: The generation phase from the interlingua
bridge meta-representation to the morphosyntax of a particular
target language will, of course, depend on the characteristics of
the target language. The generation phase implements the RRG
semantics-to-syntax linking system.
Phase 11. Generate English Morphology: The system generates English
morphology. There are some ‘tricky’ special cases, handled well
by the UniArab system, such as adding the copula verb ‘to be’ into
the English copula sentence, or changing the source language
verb’s tense to an appropriate and grammatically correct tense in
the target language, depending on the tense distinctions found
within the target language. Also, the word order in the target
language is considered and applied correctly.
Phase 12. English Sentence Generation: The process of generating an English
sentence can be as simple as keeping a list of rules, and these rules
can be extended through the life of the MT system. The system
applies some operations in English, such as vowel change in
the lexical item of English to denote sg vs. pl, for example man
vs. men. Sometimes this accompanies affixation: break/broke/broken
(¼brokeþen) to denote various tense and aspect distinctions.
Figure 21.5 Dialogue management framework for a language understanding system supporting conversational agents
and a dynamic common ground
given time, (ii) the goals that the agent will try to achieve, (iii) the actions that
the agent performs, (iv) the knowledge of the effects of these actions, (v) the
environment information the agent has (which may be incomplete or incor-
rect), (vi) the ongoing discourse interaction that the agent has with other
(human) agents and their environment over time, (vii) human language
understanding, and (viii) conversation tracking over a discourse. The parameters
of the conversational agent are characterized in its (i) personality and
emotional status, (ii) awareness of social context, (iii) model of the conversa-
tional partner (¼ other conversational actor), (iv) dialogue tracking, (v) agent
dialogue-turn tracking, and (vi) linguistic information concerning language-
specific grammar and clause structure and speech acts.
In the architecture outlined here, the relationship between agent param-
eters and behavioural choices made during the parse/generation of natural
language conversation is governed by the Agent Dialogue Manager.
Typically, this rule-based system consults other elements of the architecture
to maintain the BDI state of the conversational agent, its personality and
emotional status etc. to control a generalized natural language conversation
system and supply the appropriate parameters for conversational behaviour.
Importantly, a model of an agent system is distinguished by its active use of
social concepts.
Panesar (2017, 2019, 2020) designed and implemented this linguistically
orientated, conversational software agent framework as a proof of concept.
She discusses the relationship between natural language processing and
knowledge representation, and connects this with the goals of RRG
(Figure 21.6). In this implemented framework, Phase 1 is concerned with
Figure 21.6 Conceptual framework of the conversational software agent (from Panesar 2017: 190)
the RRG language model. Phase 2 manages the agent cognitive model and its
interfaces with the BDI model, planning model and knowledge model, while
Phase 3 handles the agent dialogue model, including dialogue management
via the RRG language model.
The advantages provided by RRG (Panesar 2019) reside in its fitness-for-
purpose for computational implementation and its level of computational
adequacy. Panesar implements a computational model of the linking algo-
rithm that utilizes a speech act construction as a grammatical object and
the (BDI) sub-model of beliefs, desires and intentions (Rao and Georgeff
1995). This work builds on elements of earlier research on the RRG
interlingua-based machine translation engine described earlier. This RRG-
based conversational software agent framework has been successfully imple-
mented in software (Panesar 2017, Pokahr et al. 2014) using conceptual
graphs and the resource description framework (RDF).
The innovation of this research resides in the combination of models and
their interoperability. The research provides new insights into the interface
between language and knowledge, via an RRG perspective.
Figure 21.8 The hand articulator in 3D space with nodes on fingers and thumb (from Murtagh
2019b: 98)
structures. The third component is the COREL scheme builder, which trans-
forms the CLS into the formal FunGramKB representation language (which
we mentioned earlier, called COREL), ultimately arriving at an extended
COREL scheme. Rules in ARTEMIS can be divided into syntactic rules to
account for the generation/recognition of the underlying layered structure
of a clause (LSC), constructional rules which guide the embedding of the
structure of argument-predicate constructions (L1 constructions in the LCM)
and lexical rules to tokenize the abstract features encoded in the LSC by
utilizing the information stored in the FunGramKB Lexicon and Ontology.
The process involved in understanding natural language with the ARTEMIS
tools is summarized in the following workflow:
Whereas RRG is able to connect the lexical entries of verbs with their
syntactic realizations via the linking algorithm, it cannot satisfactorily
account for constructional meaning. This requirement to cater for construc-
tions motivated some adjustments to the RRG model. ARTEMIS required a
constructionist linguistic model, the LCM, to be able to process this type of
meaning, since, at sentence level, argument constructions can take prece-
dence over core verbal semantics and may alter the argument structure of
the predicate. To account for such a possibility, Periñán-Pascual and Arcas
Túnez (2014) proposed and implemented in ARTEMIS a modification of the
RRG LSC by adding a new CONSTR-L1 node between the clause and the core
nodes. Cortés-Rodríguez and Mairal Usón (2016) further consider that the
addition of this new node entails the redefinition of the original RRG pre-
core slot position as a pre-C-L1 position. The rationale for this change is that
the pre-core slot may encompass not only those core constituents stated by
RRG, fronted and interrogative elements, but also constituents which are
triggered by a construction. Another necessary adjustment to the RRG
model derives from the fact that ARTEMIS shares some characteristics of
unification grammars, in such a way that parsing relies not only on syntac-
tic rules but also on the semantic and grammatical information contained
in the AVMs. A consequence of this is that, whereas in RRG abstract gram-
matical categories such as illocutionary force, aspect or negation are
described in the operator projection, in ARTEMIS these values, and the
function words associated with them, are represented in the form of
feature-bearing matrices, which now belong to the constituent projection.
In this approach, every grammatical category in the LSC has to be method-
ically described by listing the attributes that define it in a feature-bearing
matrix. At the same time, these attributes need their own description in
another AVM, in such a way that, whenever a category, an attribute or a new
part of speech (POS) is introduced, the corresponding AVM must be created
and stored in the catalogue of AVMs in the GDE. The attributes that charac-
terize each of the categories in the AVM, which would belong in the RRG
operator projection, should also in turn be defined by AVMs.
The future direction of the work of the FUNK lab, including ARTEMIS, is to
move towards a knowledge-informed intelligent cognitive agent which has
RRG as the linguistic model. This research is ongoing.
positive and crucial role to play in natural language processing. RRG delivers
a credible and realistic linguistic model to underpin the kinds of NLP
applications discussed in this chapter, and also in applications with a
functional typology orientation.
Why should a grammar deal with linking from syntax to semantics at all?
Should not specifying the possible realizations of a particular semantic
representation suffice? The answer of course is ‘no’. Functional linguists of
all persuasions refute this using the argument that theories of linguistic
structure should be directly relatable to testable theories of language pro-
duction and comprehension. Computational models assist in this process.
Based on our experience to date with computational models of RRG, we can
verify that RRG is suitable as a basis for computational implementation in
software. Computational models of RRG serve as a rich testing ground for
the linguistic theory. Indeed, Butler and Gonzálvez-García (2014: 451) note
that RRG is the only functional model in what they call ‘functional-cognitive
space’ that has a number of ongoing computational research activities.
Butler and Gonzálvez-García note in particular several of the case studies
discussed here (Guest 2008; Gottschalk 2012, 2019; Nolan 2004; Nolan and
Salem 2010, 2011; Murtagh 2019a, 2019b). In addition, recent work in
connection with the RRG-inspired FunGramKB knowledge base has seen
important advances in the computational implementation of RRG-based
procedures. Butler and Gonzálvez-García consider these studies as represent-
ing an important growing trend in RRG (see also Butler and Martín Arista
2009). An important question (Nolan 2016), from a computational RRG
perspective is: ‘What can theoretical linguistics do for natural language
processing research?’ In order to be considered fit for purpose, certain levels
of adequacy are expected from contemporary linguistic models (Butler
2009). These levels span the descriptive, typological, psychological and
explanatory adequacy required in a viable model of language. In today’s
world we can consider that another level of adequacy is also required, that
of computational adequacy. This is a concept well known to those scientists
and professionals working within computer science and software develop-
ment. A computer system and its underlying model must be fit for purpose
as specified. It must deliver its results in a coherent, timely and efficient
manner with efficient utilization of available resources. In linguistic model-
ling terms, a linguist is concerned with the model of the computation in the
mind of a person who actually computes the link between the concept–
semantics–syntax interfaces in an utterance. In computational linguistic
terms, however, this model is then a subset of the model of grammar used,
and once implemented in software, its systematic behaviour. Specifically,
considered as a computational system, does the model work to deliver the
correct and expected results optimally and efficiently? An implication of the
requirement of computational adequacy for a linguistic model imple-
mented in software is that one must address the appropriate levels of
granularity required for software specifications in order to actually model
a grammar. This extra level of detail needed to specify the model of gram-
mar such that it is understandable by a computer system puts additional
and complex demands on the linguist to reach the correct levels of granu-
larity and formalized precision so that the model can be programmed.
For RRG, where syntax is not viewed as autonomous but semantically
motivated, the scale of the challenge is increased. However, as a scientific
enterprise, it is worthwhile as it has the potential to model the computa-
tion of language from lexicon, semantics and morphosyntax. In function-
ally motivated models of language a grammar–lexicon–construction
continuum exists and the interfaces between these need to be rigorously
expressed (Nolan and Periñán-Pascual 2014; Nolan and Salem 2011; Nolan
2014). As a way of treating linguistic and computational complexity in
pursuit of linguistic realism with a robust cultural awareness of linguis-
tic conventions, this scientific enterprise of creating linguistically
motived language software and computational models of grammar is
crucial. The combination of computer science with RRG-style linguistics
has the potential to create many useful innovations that will cause a
phase shift in our expectations (Nolan and Diedrichsen 2019; Kailuweit
et al. 2019). As it has become increasingly pervasive in our lives, human
language technology is now a central component of computer science
and computational linguistics in its treatment of natural language in
software. Central to future research and development in this area is the
convergence of knowledge of computer science and linguistics and other
related disciplines. Crucially, the key ingredients are an understanding of
linguistics and how languages work to meet the levels of adequacy
needed to be successful and accepted, and to guide computer science in
application development and the engineering of solutions of value to us.
The functionally oriented RRG theory of linguistics can contribute con-
siderably to NLP research.
One way to measure a syntactic theory in practical computational appli-
cations is to consider the computational complexity of its implemented
parsing and generation algorithms. At its present stage of development,
computational work in RRG has tended not to address issues relating to
computational complexity at a theoretical or empirical level. To date, the
focus of computational work has been on the creation of RRG-based soft-
ware applications that serve some particular purpose. Consequently, com-
putational work on RRG has not focused theoretically on the science of
parsing, concentrating instead on implementing linguistically motivated
parsers for particular applications and a variety of languages. One excep-
tion to this is found in Kallmeyer et al. (2013), which describes a body of
research on the computational formalization of RRG that uses the building
blocks provided by the theory of Tree Adjoining Grammar (TAG), with
modified operations for combining trees. In this, Kallmeyer et al. (2013:
175) report on a tree rewriting system, called Tree Wrapping Grammar
(TWG), that captures the basic tree-composition principles of RRG, with the
References
Fumero Pérez, M. and A. Díaz Galán. 2017. The interaction of parsing rules
and argument-predicate constructions: Implications for the structure of
the Grammaticon in FunGramKB. Revista de Lingüística y Lenguas Aplicadas
12: 33–44.
Gottschalk, J. 2012. The persuasive tutor: A BDI teaching agent with Role and
Reference Grammar language interface – Sustainable design of a conver-
sational agent for language learning. ITB Journal 13(2): 31–50. https://arrow
.tudublin.ie/itbj/vol13/iss2/3.
Gottschalk, J. 2014. Three-place predicates in RRG: A computational
approach. In Nolan and Periñán-Pascual (eds.), 79–104.
Gottschalk, J. 2019. On the application of conceptual graphs in RRG: First
steps towards a functional computational processing model. In R. Kailu-
weit, L. Künkel and E. Staudinger (eds.), Applying and Expanding Role and
Reference Grammar, 325–354. Freiburg: NIHIN Studies. https://freidok.uni-
freiburg.de/data/16830.
Guerra García, F. and E. Sacramento Lechado. 2014. Exploring the thematic-
frame mapping in FunGramKB. In Nolan and Periñán-Pascual
(eds.), 233–250.
Guest, E. 2008. Parsing for Role and Reference Grammar. In Van Valin (ed.),
435–453.
Jackendoff, R. S. 1990. Semantic Structures. Cambridge, MA: MIT Press.
Kailuweit R, L. Künkel and E. Staudinger (eds.). 2019. Applying and Expanding
Role and Reference Grammar. Freiburg: NIHIN Studies. https://freidok.uni-
freiburg.de/data/16830.
Kallmeyer, L., R. Osswald and R. D. Van Valin, Jr. 2013. Tree wrapping for Role
and Reference Grammar. In G. Morrill and M.-J. Nederhof (eds.), Proceedings
of the Formal Grammar 17th and 18th International Conferences, 175–190. Hei-
delberg: Springer.
Kecskes, I. and F. Zhang. 2009. Activating, seeking, and creating common
ground: A socio-cognitive approach. Pragmatics & Cognition 17:(2): 331–355.
Leeson, L. and B. Nolan. 2008. Digital deployment of the Signs of Ireland
Corpus in Elearning. In Language Resources and Evaluation LREC2008 – 3rd
Workshop on the Representation and Processing of Sign Languages: Construction
and Exploitation of Sign Language Corpora, Marrakech, Morocco.
Mairal Usón, R. and F. Cortés-Rodríguez. 2017. Automatically representing
text meaning via an interlingua-based system (ARTEMIS): A further step
towards the computational representation of RRG. Journal of Computer-
Assisted Linguistic Research 1: 61–87.
Mairal Usón, R. and F. Ruiz de Mendoza. 2009. Levels of description and
explanation in meaning construction. In C. S. Butler and J. Martín Arista
(eds.), Deconstructing Constructions (Studies in Language Companion Series
107), 153–198. Amsterdam: John Benjamins.
Montiel-Ponsoda, E. and G. Aguado-de-Cea. 2014. Applying the lexical con-
structional model to ontology building. In Nolan and Periñán-Pascual
(eds.), 313–338.
publication/319376849_PhD_Thesis_%27A_linguistically_centred_text-
based_conversational_software_agent%27.
Panesar, K. 2019. Functional linguistic based motivations for a conversa-
tional software agent. In Nolan and Diedrichsen (eds.), 340–371.
Panesar, K. 2020. Conversational artificial intelligence: Demystifying statis-
tical vs. linguistic NLP solutions. Journal of Computer-Assisted Linguistic
Research 4: 47–79.
Periñán-Pascual, C. 2013. Towards a model of constructional meaning
for natural language understanding. In Nolan and Diedrichsen (eds.),
205–230.
Periñán-Pascual, C. and F. Arcas-Túnez. 2004. Meaning postulates in a lexico-
conceptual knowledge base. In Proceedings of the 15th International Workshop
on Databases and Expert Systems Applications, 8–42. Los Alamitos, CA: IEEE.
Periñán-Pascual, C. and F. Arcas-Túnez. 2005. Microconceptual-knowledge
spreading in FunGramKB. In Proceedings of the 9th IASTED International
Conference on Artificial Intelligence and Soft Computing, 239–244. Anaheim:
ACTA Press.
Periñán-Pascual, C. and F. Arcas-Túnez. 2007. Deep semantics in an NLP
knowledge base. In Proceedings of the 12th Conference of the Spanish Association
for Artificial Intelligence, Universidad de Salamanca, 279–288.
Periñán-Pascual, C. and F. Arcas-Túnez. 2008. A cognitive approach to qual-
ities for NLP. Procesamiento del Lenguaje Natural 41: 137–144.
Periñán-Pascual, C. and F. Arcas-Túnez. 2010. The architecture of Fun-
GramKB. In Proceedings of the International Conference on Language Resources
and Evaluation, LREC 2010, Malta, 2667–2674. www.lrec-conf.org/proceed
ings/lrec2010/summaries/284.html.
Periñán-Pascual, C. and F. Arcas-Túnez. 2014. The implementation of the Fun-
GramKB CLS Constructor in ARTEMIS. In Nolan and Periñán-Pascual (eds.),
165–196.
Periñán-Pascual, C. and R. Mairal Usón. 2009. Bringing Role and Reference
Grammar to natural language understanding. Procesamiento del Lenguaje
Natural 43: 265–273.
Periñán-Pascual, C. and R. Mairal Usón. 2010. Enhancing UniArab with
FunGramKB. Procesamiento del Lenguaje Natural 44: 19–26.
Periñán-Pascual, C. and R. Mairal Usón. 2011. The COHERENT methodology
in FunGramKB. Onomázein 24: 13–33.
Pokahr, A., L. Braubach, C. Haubeck and J. Ladiges. 2014. Programming BDI
agents with pure Java. In German Conference on Multiagent System Technologies,
MATES2014, 216–233. Switzerland: Springer International Publishing.
Pustejovsky, J. 1991. The generative lexicon. Computational Linguistics 17(4):
409–441.
Pustejovsky, J. 1995. The Generative Lexicon. Cambridge, MA: MIT Press.
Rao, A. S. and M. P. Georgeff. 1995. BDI agents: From theory to practice.
In Proceedings of the 1st International Conference on Multi-Agent Systems ICMAS-
95, 312–319.
Notes
Abbreviations
22.1 Introduction
* The data come mainly from my native consultants, supplemented by existing language materials, such as two collections of
texts (Leman 1980a and 1987) and a dictionary (Fisher et al. 2006). I wish to express my gratitude to the Cheyenne people,
especially my late friend Ralph Redfox, for kindly sharing their knowledge of this language with me. Needless to say, all
errors remain my sole responsibility.
Salience IRR TNS DIR (2) E. MOD EQ *NEG D. MOD ASP DIR (1) initial+medial+final Voice *NEG Cross-reference EVID IF
Figure 22.1 Verb structure in Cheyenne (modified from Corral Esteban 2017: 310)
as they cross-reference the core arguments of the predicate and cover gram-
matical information about them in terms of person, number, animacy, and
proximate/obviative status.
The two outermost operators, the clausal operators of evidentiality and
illocutionary force, follow the pronominal suffixes. To summarize, in Chey-
enne operators realized as bound morphemes occur on both sides of the
nucleus.
Example (1) and its representation in Figure 22.2 indicate that, in Chey-
enne, the order of the morphemes expressing operators with respect to the
nucleus reflects their relative scope and, consequently, confirms the validity
of the principle governing this ordering (Van Valin and LaPolla 1997: 49–52),
SENTENCE
CLAUSE
CORE
PRED
DIR NUC
ASP NUC
MOD CORE
TNS CLAUSE
CLAUSE EVID
CLAUSE IF
SENTENCE
Figure 22.2 Layered structure of the clause with constituent and operator projections
In these two transitive constructions, the verb has two semantic arguments –
a second-person singular argument and a third-person singular argument.
As second person outranks third person in the Person Hierarchy, both
constructions select the prefix né- ‘2’, regardless of the semantic function
of the argument in each construction. However, despite having the same
prefix, each construction exhibits different relations between the verb and
its arguments, as indicated by the suffixes -o and -a, which reflect a direct
and an inverse construction respectively. In (2), the person cross-referenced
by the prefix né- indicating a second-person participant, corresponds to the
higher semantic macrorole, namely the actor, so the Person and Semantic
Function hierarchies are in proper alignment, and the construction is then
marked as direct. In (3), by contrast, the hierarchies are not properly aligned,
as the same person prefix né- corresponds to the lower semantic macrorole,
that is to say, the undergoer, so the construction is marked as inverse.
The main difference between the two verbal orders lies in the function of
their prefix: whereas in the Independent order the prefix serves to express
the most pragmatically salient person, in the Conjunct order it indicates the
grammatical mood, namely realis (e.g. indicative tsé- (5–7)) or irrealis (e.g.
conditional vé’- (8), optative momȯxe- (9), etc.). The prefix in a dependent
clause functions as a clause linkage marker and is usually accompanied by
a preverbal particle that identifies the type of subordinate clause (e.g. -heše-
in complement clauses (5), -éše- in temporal adverbial clauses (6), -homá’xe- in
adverbial clauses of reason (7), etc.), which can lead to an instance of either
daughter subordination (5) or ad-subordination (6–8).
(16) É-ho’ééto-Ø.
I-snow.vii-Isg.a
‘It is snowing. (lit. ‘It snows.’) snow′(Ø)
(18) Ná-mané-me.
1-drink.vai-1pl.a
‘We drink.’
do′ (1pl, [drink′ (1pl, Ø)])
The activity verb mane ‘drink’ in (18) has one direct core argument that is coded
on both the prefix ná- ‘1’ and the pronominal affix -me and cross-references a
first-person plural animate argument, namely the first argument in the LS.
However, no reference is made to the second semantic argument of the verb;
hence the verb mane ‘drink’ in (18) can be said to take one macrorole, which is
assigned to the only direct core argument of the verb.
Arguments with generic reference are generally cross-referenced on the
verb in Cheyenne.
In this example, both the prefix é- ‘3’ and the pronominal affix -ovo cross-
reference two arguments – a third-person/proximate plural animate argu-
ment and a fourth-person/obviative singular argument. However, despite
being cross-referenced on the verb, the two arguments are not macroroles
since they are non-specific.
There is, however, a context where an argument with a generic reference
is not coded on the verb in Cheyenne. This occurs when there is a close
semantic relationship between the verb and the argument, which would be
a typical object of the verb. In such cases, the object RP is incorporated into
the verb.
(20) a. É-o’ene-mēna-Ø.
3-pick.in-berry.fai-3sg.a
‘He is berry picking.’ (Fisher et al. 2006: 215)
do′ (3sgM, [pick′ (3sgM, o’ene)])
b. É-o’en-anȯtse menȯtse.
(3) -pick.vti-3>I.3sg.a.Ipl.u berry.pl
‘He is picking the berries.’
do′ (3sgM, [pick′ (3sgM, menȯtse)]) & ingr picked′ (menȯtse)
The difference between activities and their telic version, that is to say active
accomplishments, is observed syntactically in Cheyenne. Unlike the active
accomplishment transitive verb o’en ‘pick’ in (20b), the activity intransitive
verb o’enemēna ‘berry-pick’ in (20a) is formed by the initial stem of the verb
‘pick’, o’ene, and a final stem of the noun ‘berry’, mene. Also, while o’en ‘pick’
has two direct core arguments, cross-referenced by the prefix é- ‘3’ and the
suffix -anȯtse, which indexes a third-person singular animate agent and a
plural inanimate patient, the bipartite verb stem o’ene-mēna ‘berry-pick’ has
only one direct core argument – a third-person singular animate agent, which
is coded by the prefix and the null suffix. Additionally, as only the first
argument of the activity verb in (20a) is specific, the verb only takes one
macrorole. Conversely, the active accomplishment verb in (20b) has two
specific arguments and, therefore, takes two macroroles.
Except for these examples showing the influence of referentiality, the
number of direct core arguments and macroroles also tends to coincide in
Cheyenne transitive constructions provided the arguments are referential.
Consequently, two macrorole verbs are usually transitive.
The two-place verb ono’átam ‘respect’ has two arguments, namely a first-
person singular animate experiencer and a third-person plural animate
theme, which are cross-referenced on the verb by the prefix ná- ‘1’, the theme
marker -o- and the pronominal affix -o’o. The second argument is also
realized by a full lexical RP, namely tsé-héstoo’ éšeehaese ‘my parents’ (lit.
‘those who raised me’). As both direct core arguments are referential, the
verb can be said to have two macroroles.
Finally, unlike other Algonquian languages (e.g. Blackfoot, Plains Cree or
Ojibwa), Cheyenne ditransitive verbs have a very complex system of suffixes
(Corral Esteban 2014: 419–422), which vary with respect to the animacy and
number of the theme argument. Thus, verbs such as the stems mét ‘give’,
véestomev ‘ask for’, nomáhtseh ‘steal’, véstomev ‘promise’, or vóo’seh ‘show’
behave like true ditransitive verbs, since they cross-reference the three core
arguments morphologically.8
(26) Ná-méa(‘e)-ēme.
1-give.vai-1pl.a
‘We give something to somebody.’
[do′ (1pl, Ø)] cause [become have′ (Ø, Ø)]
the pronominal affixes -no-ne-o’o in (24) code only a first-person plural agent
and a third-person plural theme nanésonėhaneo’o ‘our children’; the pronom-
inal affixes -á-nó-ne in (25) code only a first-person plural agent and the
inanimate theme nėstámane ‘our food’; and finally, the pronominal affix -ēme
in (26) only cross-references the first-person plural agent.
The role of referentiality is not only noticed on the pronominal affixes, but
also in the verb stem, which adopts different forms depending on the
number of macroroles the verb takes. Thus, in Cheyenne, the predicate ‘give’
can be expressed through a ditransitive stem mét, a TA stem méa’tov, a TI
stem méa’e, or an AI stem méa’e (23–26). As regards the correlation between
the number of direct core arguments and macroroles, in the intransitive
verb in (26) the only direct core argument of the verb is a macrorole. The
same correlation holds in (24) and (25), as the two transitive verbs have two
direct core arguments and both are macroroles. Since only two macroroles
are allowed in RRG, in the ditransitive construction in (23) only two of the
direct core arguments are macroroles – the first-person plural agent and the
recipient nanésonėhaneo’o ‘our children’, the other – the theme nėstámane ‘our
food’ – being the non-macrorole core argument. They are all identified
through cross-referencing.9
To summarize, except for cases such as constructions including
‘weather’ verbs10 and impersonal constructions, and the problem posed
by generic reference, which can be coded on the verb via either pronom-
inal affixes or noun incorporation, M-transitivity closely corresponds to S-
transitivity in Cheyenne, due to the effect of referentiality on the coding of
arguments and, most importantly, on the form of the syntactic core tem-
plate: the presence of a non-referential argument in a transitive construc-
tion affects not only the pronominal affixes but also the form of the verb
stem, since the non-referential argument is not cross-referenced on the
verb and is, therefore, not represented within the core. Finally, the Chey-
enne examples also show that the number of macroroles that a verb takes
is less than or equal to the number of semantic arguments in its logical
structure.11
A particularly challenging case is a typical Algonquian transitive construc-
tion, commonly referred to as the ‘unspecified actor’ construction (Hockett
1996), as it includes no reference to the agent of the action denoted by
the verb.
Despite the fact that these sentences include the same arguments and are
semantically equivalent, the arguments show a different proximate/obvia-
tive status. Thus, whereas in (28) the agent is the proximate náhe ka’ėškóne
‘that boy’, and the benefactive is the obviative néhe ma’háhkėseho ‘this old
man’, in (29), there is a shift with respect to the proximate marking. Now,
the agent náhe ka’ėškóneho ‘that boy’ is obviative and the benefactive néhe
ma’háhkése is proximate.
A representation of direct core arguments (28) is provided in Figure 22.3
for a subsequent comparison with that of incorporated arguments in appli-
cative constructions in Figure 22.4 and argument-adjuncts and adjuncts in
Figure 22.5.
As there is no correlation between the proximate/obviative distinction and
semantic roles or syntactic functions, it is necessary to take into account the
information provided by the nominal morphology and the prefix and pro-
nominal suffixes on the verb. Thus, the difference between these two
examples of transitive predication lies in that, while in the direct construc-
tion in (28) the proximate argument náhe ka’ėškóne ‘that boy’ is the actor and
the obviative argument néhe ma’háhkėseho ‘this old man’ is the benefactive,
the opposite occurs in the inverse construction in (29), as the obviative
argument náhe ka’ėškóneho ‘that boy’ is now the actor and the proximate
argument néhe ma’háhkése ‘this old man’ is the benefactive.
SENTENCE
CLAUSE
CORE
CORER CORER
N V N
Náhe ka´ škónei éi- ohk - vést hém - óhoi+j néhe ma´háhk sehoj
SENTENCE
CLAUSE
CORE PERIPHERY
P V N N
SENTENCE
CLAUSE
CORE PERIPHERY
P V N N N
The locative adjunct méóne ‘on the road’ is formed by the noun méó’o ‘road’
plus the oblique nominal suffix -e, which denotes location. This locative
adjunct is neither cross-referenced on the verb nor incorporated into the
verbal complex as a stem-forming element.
Despite their reference to one obligatory argument of a verb, argument-
adjuncts in Cheyenne are not cross-referenced on the verb, which makes
them resemble adjuncts syntactically. Cheyenne also does not distinguish
adjuncts from argument-adjuncts in terms of marking, as the same oblique
CORE
PRED
SENTENCE
PARSER
CORE
1
PRED
This linking procedure from semantics to syntax starts with the construc-
tion of the semantic representation of the sentence, which is drawn from
the logical structure of the predicate.
As is the case with other languages, the selection of the actor and under-
goer arguments in monotransitive constructions presents no difficulty in
Cheyenne, as the highest-ranking argument in the logical structure is
selected as the actor and the lowest-ranking argument is selected as the
undergoer. Ditransitive constructions indicate that Cheyenne represents
an instance of what Dryer (1986) called a ‘primary object language’, that is
to say a language where the only pattern that occurs with three-argument
predicates corresponds to the marked selection for undergoer.15 Thus,
when a verb has three arguments, the second-highest, rather than the
lowest-ranking argument in the logical structure is selected as the under-
goer macrorole and, consequently, the lowest-ranking argument is the
non-macrorole direct argument. In example (33), therefore, the leftmost
argument in the logical structure of na-né’ame ‘my parents’ is selected as
the actor, and the second leftmost argument – the participant referring to
the speaker – is chosen as the undergoer instead of the expected rightmost
argument mȯxe’ėstoo’o ‘book’, which now becomes the non-macrorole
argument.
The determination of the morphosyntactic coding of the arguments shows
an interesting specific feature of Algonquian languages, as the choice of a
specific argument on the prefix shows the existence of a privileged syntactic
argument (PSA) in these languages. This PSA involves a pragmatically deter-
mined neutralization of semantic roles for syntactic purposes, as the choice
of the prefix is pragmatically determined – it is only determined by the
Person Hierarchy – and can correspond to different semantic roles
depending on the ranking of the participants on the hierarchy (e.g. agent
(4), patient (19), experiencer (1), recipient (33), etc.). In a direct transitive
construction, the PSA corresponds to the actor, and in an inverse transitive
construction the PSA matches the undergoer. Thus, in example (33), bearing
in mind that it is an instance of inverse construction, the PSA is the under-
goer cross-referenced on the prefix ná- ‘1’.
As regards the syntactic representation of the sentence, the syntactic
template for a three-place predicate like mét ‘give’ must contain a core
including three argument positions, which will be filled by the pronominal
arguments cross-referencing the core arguments, namely the prefix ná- ‘1’
and the portmanteau verbal suffix -aenóvo, which cross-references the three
semantic arguments of the verb.
Finally, the linking of arguments to positions in the syntactic template is
also an interesting feature of Cheyenne. The fact that word order in this
language is pragmatically determined allows for greater flexibility when it
comes to selecting the syntactic template of the sentence because of the
positions (e.g. pre-core slot (PrCS) or any extra-core slot (ECS)) to which the
independent RPs are assigned.
The direction of the linking process from syntax to semantics involves the
interpretation of the overt morphosyntactic form of a sentence and the
deduction of the semantic functions of the elements in the sentence deriv-
ing from such an interpretation, as shown in Figure 22.7.
The information provided by the verbal affix -aenóvo reflects the fact that,
in this inverse construction, the verb is ditransitive and has three direct core
arguments – a third-person plural agent, a first-person singular recipient,
and an inanimate singular theme. The prefix also enables us to identify
which constituent acts as the PSA, since, according to the ranking of the
Person Hierarchy, which favours first person over third and inanimate
person, ná- ‘1’ acts as the PSA of the construction. As regards the identifica-
tion of the semantic macroroles of the core arguments, a ditransitive verb
like mét ‘give’ has three possible candidates for only two macroroles. How-
ever, taking into account the accurate grammatical information provided by
the bound markers ná- ‘1’ and -aenóvo, and considering that Cheyenne
invariably shows the marked undergoer selection in ditransitive construc-
tions, we can then select the proximate participant na-né’ame ‘my parents’ as
actor (‘x’) and the first-person participant as undergoer (‘y’), whereas the
third and last core argument, namely the inanimate participant mȯxe’ėstoo’o
‘book’, becomes the non-macrorole argument (‘z’).
The linking of the core arguments to the corresponding slots in the semantic
representation of the sentence may be problematic in Cheyenne due to the fact
that the number of pronominal affixes in the verb does not necessarily corres-
pond to the number of direct core arguments. This is mainly because the
verbal suffix is generally a portmanteau that cross-references all the direct
core arguments of a verb, and the prefix only codes one of these core argu-
ments, which does not allow for a one-to-one correspondence between direct
core arguments and semantic arguments. Thus, it is necessary to bear in mind
what type of stem the verb presents and what kind of grammatical infor-
mation is provided by each of the pronominal affixes in order to gauge its
number of direct core arguments. It would therefore seem logical to assume
that the portmanteau cross-references all the arguments and the prefix cross-
references one of them again, as represented in Figure 22.7.
The linking algorithm shows the important role played by pragmatics in the
interaction between semantics and syntax in Cheyenne and reveals a series of
grammatical properties: the variable word order exhibited by RPs, the prag-
matically determined PSA, the cross-referencing of arguments through a port-
manteau, and even the lack of correlation between the proximate/obviative
status of referents and their semantic function in non-local contexts.
SENTENCE
CLAUSE
PrDP PoDP
CORE
PrCS RP (RP)
PRO NUC PRO
PRED
V
Topic Shifting/Information/Contrastive Verbal complex Information/Non-shifting Familiar
Part/Adv Topic Focus Topic/Focus Focus Topic Topic
NEWSWORTHY NON-NEWSWORTHY
Figure 22.8 Information structure-based sentence template
SENTENCE
CLAUSE PoDP
CORE RP
V
Éi-sáa-méev-ó- he – hoi+j neše mo´éhno´hamej (Richard)
SENTENCE
CLAUSE
CORE
PRO NUC RP
PRED
V
é-onén šeotse-Ø na-am ho´héhame
SENTENCE
CLAUSE
PrCS CORE
RP PRO NUC
PRED
V
Na-amaho´héhame é-onén še-otse-Ø
Na-am ho´héhame
Figure 22.11 Unmarked narrow-focus structure in Cheyenne
The most common word order pattern in this situation, which places the
verb in clause-initial position and the RP in postverbal position, represents
the unmarked structure of this focus type because there is no element
occupying the PrCS slot. Figure 22.10 represents the focus structure of (35).
As there is no presupposition in this structure, the assertion extends over
the entire proposition, as is indicated by the actual focus domain in
Figure 22.10. Both the core and the subject RP in the ECS are associated
with special intonational realization, although its intensity decreases as it
approaches the clause-final position.
distinction between singular and plural number (see examples 11 and 13) or
obviative marking (see 15). However, in transitive constructions, where more
than one participant is involved, the prefix only cross-references the most
pragmatically salient argument in terms of the Person Hierarchy.
(39) Né-vóom-ȧtse.
2-see.vta-1>2.1.sg.a.2sg.u
‘I saw you.’
(40) Né-vóom-e.
2-see.vta-2>1.2sg.a.1sg.u
‘You saw me.’
In these examples the prefix né- ‘2’ is selected as the prefix heading the verbal
complex, thereby acquiring a special syntactic status, namely the PSA of the
construction. Neither the prefix né- ‘2’ nor the pronominal suffix -ȧtse or -e
can determine the semantic roles of the arguments of the verb by them-
selves. The prefix cross-references only the higher-ranking person, but is not
indicative of semantic role on its own – it cross-references the undergoer in
(39) and the actor in (40) – and the portmanteau formed by the theme
marker plus the pronominal affixes tends to provide information about
the direction of the construction, and to cross-reference the direct core
arguments of the verb, but is sometimes ambiguous, especially in the
Conjunct order. An examination of both elements, therefore, becomes neces-
sary in order to link arguments and semantic roles. Thus, the prefix in both
examples is a second-person participant; the portmanteaus -ȧtse and -e indi-
cate that (39) and (40) are an inverse and a direct construction respectively,
and that these constructions include a first-person singular actor and a first-
person singular undergoer, respectively.
The direct/inverse mechanism is especially useful when the point of view
changes for pragmatic reasons, and the core arguments are not lexicalized
as RPs.
22.4 Conclusion
The goal of this grammatical sketch has been to explore sentence structure
in Cheyenne from the perspective of RRG through the analysis of a range of
grammatical issues serving both to reveal the core components shared by all
languages and to highlight those that are specific to Cheyenne grammar.
The analysis of Cheyenne grammar in Section 22.2 showed that there is no
evidence for the postulation of grammatical relations in addition to seman-
tic predicate–argument relations, save for a pragmatically influenced PSA
that is represented by the verbal prefix. It also indicates the hierarchical
scope order of operators in Cheyenne, which is broadly defined in accord-
ance with the RRG proposal. Finally, it reveals the fundamental role played
by pragmatics, especially in terms of argument coding, macrorole assign-
ment and word order.
Section 22.3 argued that, despite the word-order variability displayed by
Cheyenne, it is possible to integrate information structure into clause struc-
ture and explore the intricate mechanism used by this language to accom-
modate semantic information into syntactic structure. Cheyenne links
arguments and particular semantic roles through an intricate mechanism
consisting of two components, which work in combination with the prox-
imate/obviative distinction in contexts involving non-local participants.
These components are illustrated by: (1) the correlation between the Person
Hierarchy and a Semantic Function Hierarchy and (2) a binary system of
verbal direction expressed by a theme marker or direction-marking mor-
pheme on the verb, which is complemented by the grammatical informa-
tion about the participants provided by the pronominal affixes. The direct/
inverse system works along with the reference-tracking system of obviation,
References
Notes
1 The prefix selecting the most pragmatically salient person only occurs in the
Independent order (except in the conjectural mode, which is introduced by
the prefix mó-). The prefix in the Conjunct order conveys other grammatical
meanings (e.g. tsé- ¼ realis, vé’- ¼ irrealis, mȧh- ¼ potential, hó’- ¼ iterative,
momóxe- ¼ optative, etc.). The independent and the conjunct verbal orders
generally occur in main and subordinate clauses, respectively.
2 There are two main types of directional particle: (1) the cislocative neh- and
translocative ta-, which serve to express the locational or temporal deixis
of the action, with the speaker or the speech act itself as the deictic locus;
and (2) phoric particles, such as the anaphoric and cataphoric particles nė-
and tsė-, which serve to link elements in discourse, namely a place old or a
place new in discourse respectively, and appear close to the prefix. There
are also other directional initial particles such as he’am- ‘up’, e’e- ‘up’, an
(ȯhe)- ‘down’, ést- ‘into’, sé’- ‘into’, hóe- ‘out’, etc., which always occupy the
rightmost position preceding the verb.
3 Negation in Cheyenne is commonly expressed through two separate and
complementary particles, namely sáa-, a preverbal particle, and -hé, a
postverbal particle that normally occurs between the theme marker and
the pronominal affixes.
4 For the sake of simplicity and because it is not always possible to separate
the theme marker from the pronominal affixes on a verb, both parts will
be analysed as a portmanteau suffix throughout the chapter. In order to
indicate the direction of constructions including transitive verbs, a nota-
tion such as ‘1>3’ will be used, for example to indicate a first-person actor
and a third-person undergoer. For intransitive verbs, its only argument
will be referred to as ‘1’, for example to indicate that the actor or under-
goer is the first person.
5 Adjectives in Cheyenne are always realized as bound particles that
attach to verbs (see example 38(6), in section 22.3.2) or nouns (11) and
behave like adverbs or adjectives in English, respectively. Likewise, they
can also form non-verbal predicates when they are attached as initials
to final verb stems in attributive constructions (see 13). By contrast,
pronouns (see 37(4)), numerals (see 34), adpositions (see 37(4)), and
some adverbs (see 37(3)) are realized as free particles in Cheyenne
and, consequently, excluded from the possibility of functioning
as predicates.
6 Albeit less frequently, other types of identificational sentence can be
expressed in Cheyenne by means of a linking or copular verb (a), the
combination of a preverbal particle and a nominal predicate (b), or even
a verb-less clause (c).
The AI verb penȯha ‘grind’ only codes a first-person singular agent and an
inanimate plural patient.
14 It is very difficult to come to a decision about the status of the suffix -va.
On the one hand, given there is an applicative (incorporated adposition)
licensing hoo’hénóva ‘in the bags’ in (32), it does not seem likely that this
applied argument may be marked by a postposition. On the other hand,
the fact that the suffix -va may occur on a variety of argument-adjuncts
and adjuncts with different meanings (e.g. time, location, instrument)
appears to suggest that it is not a case marker in the usual sense. In view
of the above, it seems plausible to consider it a kind of generalized
oblique marker, with no inherent semantics of its own.
15 The evidence for the marked undergoer choice in ditransitive construc-
tions comes from both coding (indexing and ordering of bound forms)
and behavioural properties (reflexivization, reciprocalization, and direct/
inverse marking) (Corral Esteban 2014: 403–415). On the one hand, the
ditransitive verbal paradigms resemble their TA more than their TI
counterparts, and the pronominal markers cross-referencing the recipi-
ent occur closer to the verb than those cross-referencing the theme. On
the other hand, the reflexive/reciprocal suffix only permits coreference
between actor and recipient, but not between actor and theme, and
direct/inverse marking shows a clear secundative bias since the impossi-
bility of constructing ditransitive sentences with inanimate recipients
that are prototypically human, or first- or second-person themes that are
prototypically non-human, implies that only the person of either the
actor or recipient, but not that of theme, can appear as a prefix.
16 I am assuming that focal elements occupy the PrCS rather than the PrDP
in Cheyenne because there is no intonation break between most of the
preverbal RPs and the verbal complex, which suggests that they are part
of the clause.
17 Albeit infrequently, it is also possible to find left-dislocated familiar
topics in Cheyenne.
18 This assumption concerning word order is based on my analysis of word
order patterns in independent clauses found in thirty-five Cheyenne
stories told by different native speakers, which were randomly taken
from Leman (1980a). The analysis revealed that the most common sen-
tence type involves no explicit RPs, and the second most common type
involves one RP. The analysis of intransitive clauses including an RP
shows that the most common word-order pattern is SV (63.48%), which
appears to be more related to newsworthiness (new discourse topic:
83.01% vs. continuing topics: 16.99%) than to givenness (new informa-
tion: 59.47% vs. old information: 40.53%). In transitive clauses including
two overt RPs, the most common word-order pattern is SVO (86.67%),
followed far behind by OVS (8.89%). In the more common situation
displayed by transitive clauses, that is involving an object RP, the most
common pattern is VO (66.41%)).
19 This situation also appears to hold in presentational sentences, where all
the information is asserted and nothing is taken as presupposed. While
Cheyenne appears to follow a verb-initial pattern as its basic, but not
obligatory, word order, sentences that include a direct core argument in
preverbal position also appear to be very common at the beginning of a
story or conversation, because of the speaker’s interest in establishing a
discourse topic from the outset.
Abbreviations
MOD-PRO-ASP-ADV-ELEV/DIR-VAL-NINCORP-V-VAL-DIR-ASP-TNS-PC-PRO
end (and again usually the former), but this is only a tendency, and it is
not unusual to find them midway in a clause, say, following a noun
functioning as a core argument.
In any case, Yimas clauses, like those of many Papuan and other purely
oral languages, tend to be information-poor, containing few constituents.
Clauses containing a verb and two overt core argument nouns are very rare
indeed, much less than 1 per cent of all clauses. Where they might occur is
at the beginning of a narrative to introduce the characters. Yimas has no
specialized presentative constructions for introducing new referents like
‘there was an old woman who lived . . .’, so new nouns are introduced baldly
in new mentions at the beginning of a story, and this usually by naming
them; ‘Yapalmay and Mampalmay lived with their brother Yampwiŋkawi’,
for instance, is the opening line of the legend which explains how men’s
penises got shortened. There is no distinct clause type for sentence focus.
About half of all Yimas clauses with transitive verbs consist of just the verb
with perhaps a peripheral temporal or locative expression (often marked
with the oblique suffix -n ~ -nan, the only true nominal case marker in the
language), and the other half will have a single overt noun as a core
argument, again with perhaps a peripheral adjunct. What determines these
patterns is mainly information structure. Nouns whose referents are acti-
vated or easily accessible are normally only expressed by the corresponding
bound affixes for their noun class and number. To have an overt noun and
its corresponding bound affix co-occur in a context in which there is no
ambiguity as to their reference makes the referent of the noun highly
contrastive. It takes on this contrastive reading from the conflict in inter-
pretation between the overt noun and the bound affix. Overt nouns are used
when their referents are inactivated or inaccessible and thereby being
introduced into the stream of speech; they are focal in information struc-
ture, either part of predicate focus or themselves as narrow focus, as clearly
shown in this question-pair.
The combination of the overt noun and bound verbal prefix here is very
strange because it forces a contrastive reading where the question does not
set a context in which this is felicitous. The overt noun signals an inactivated
reading, while the bound verbal prefix, an activated or easily accessible one,
two contradictory readings that could only be reconciled in a contrastive
focus reading, but this makes little sense in this context. So the two most
common clause types for Yimas are those with just a verb, where the refer-
ents of all core participants are activated or accessible, or a verb and one
overt noun, in which the overt noun has an inactivated or inaccessible
referent. This is the basic system, though there are a few wrinkles and
complications, particularly with wh-questions (Foley 1991: 430–433).
Due to the sensitivity of the bound verbal affixes to information struc-
ture, the analysis of the head-marking language Lakhota in Van Valin
(2013) needs to be expanded a bit. Van Valin (2013) did not intend his
analysis of Lakhota to be applicable to all head-marking languages, but
rather focused on the type represented by Lakhota, and in this spirit we
offer this analysis of Yimas as a further enrichment of the descriptive
typology. For Lakhota, Van Valin argues that the verbal prefixes, which
may be null, saturate the argument positions within the core and that any
overt nouns are in core-external, but clause-internal positions, essentially
sisters of the core node under the clause. Given that no basic bound affix of
any person, number or noun class combination in Yimas has a null expo-
nent, it would be extremely undesirable to appeal to null affixes in this
language, but without them, to apply Van Valin’s analysis of Lakhota to
Yimas would require violating the Completeness Constraint, an even more
undesirable result, if we want to satisfy all the lexical requirements of the
verb and its subcategorized arguments in the core, as we clearly do in this
language with no voice alternations or extractions. So instead, I propose
that both bound verbal affixes and overt nouns occur in the core, as long as
the nouns are inactivated, that is, they do not co-occur with a bound prefix.
Those that do, occur in the same core-external position that Van Valin
proposes for Lakhota. All of this is illustrated in the following very typical
Yimas sentence and its representation in layered structure and informa-
tion structure. The only constraint on word order here is that the conjunc-
tion kanta ‘but’ occurs preferentially in second position, though this is not
rigid, and the verb cannot be initial; otherwise all possibilities are accept-
able, though some will be more marked pragmatically. Note particularly
the separation by this conjunction of the modifier kamta- ‘empty’, which is
actually a nominalized verb in a relative clause (Section 23.7.2), hence the
tense marker -k irr, from the noun parwa ‘dock’ that it modifies, but linked
to by the noun class and number concordial suffix proper to nouns of this
class and number. Note too the fact that the oblique case suffix -n only
occurs on the noun parwa ‘dock’, but applies to its linked modifier kamta-
‘empty’ as well. The bound verbal affix is a circumfix for this verb because it
has negative polarity (Foley 1991: 251–263).
In Figure 23.1, which represents the structure of (4), RX stands for referential
expression, in order to treat nouns and bound prefixes with a single label
SENTENCE
CLM CLAUSE
RXIX.SG
FOCUS
V (INACT)
NUCLEUS
CORE
SENTENCE
Figure 23.1 Clause structure with constituent, operator and focus projections
and avoid positing phrases where there are none, as such are very depaup-
erate in Yimas. I have also simplified for purposes of exposition the
representation of the relative clause, as these will be discussed in more
detail in Section 23.7.2, and treated it like an adjective, as here it behaves
exactly as a true adjective like kpa ‘big’ would.
The syntactic structure of phrases is extremely simple in Yimas, and what
look like noun phrases are more akin to compounds in structure than
phrases. Essentially a noun phrase can consist of no more than two constitu-
ents, a noun and its preposed modifier, and those modifiers can only be of
two kinds, a possessor, which must be a pronominal or a proper name, or
one of the three adjectives, and these cannot co-occur. We can say ama-na kay
1sg-poss canoe ‘my canoe’ and kpa kay big canoe ‘big canoe’ in that fixed
order of constituents, but a noun phrase like *ama-na kpa kay 1sg-poss big
canoe is impossible in Yimas. One would say ama-na kay kpa-y 1sg-poss
canoe.viii.sg big-viii.sg ‘my canoe, the big one’, where the adjective kpa
‘big’ has been converted into a noun by the noun class and number con-
cordial suffix -y and, like kamta-k-wa empty-irr-ix.sg ‘empty’ in (4), can be
floated away from the noun it modifies and no longer forms a constituent
with it, so that this is actually two referential phrases in apposition to each
other. Ignoring nominalizations of verbs to be discussed below, referential
phrases in Yimas can be made up of at most two constituents, such as ama-na
kay or kpa kay, and the vast majority consist of just one, so I will continue to
refer to them as referential expressions (RX) in preference to RPs, especially
as they are typically at the word (X0) level. The only operator that RXs take is
number, at the core level, singular, dual and plural, and adjectives and
possessives are peripheral modifiers, so, as in Figure 23.2.
RX RX
COREN COREN
ADJ N N
NUCLEUSN NUCLEUSN
RX RX
The only other phrase type in Yimas is the postpositional phrase headed by
postpositions like nampan ‘toward, for, because of’ or kantk ‘together with’,
which form the predicate and nucleus of the phrase and their core argu-
ments are RXs. Postpositional phrases are always predicative peripheral
adjuncts; there are no non-predicative peripheral adjuncts in the language,
because, as we shall see next, all arguments subcategorized by a verb must
be projected as arguments in the core, either as an independent RX or as a
bound verbal one.
Yimas, like some other Papuan languages (e.g. Kalam; Pawley 1993), has a
restricted lexicon of monomorphemic verbs, only around a couple of hun-
dred verb roots. More complex verbal expressions are constructed by these
roots in serial verb constructions in nuclear junctures. Yimas has basic
underived intransitive, transitive and ditransitive verbs, although there
are only four roots in the last class. There are also lexical processes that
derive verbs, a reciprocal prefix that removes a core argument and two
causative prefixes, six applicative affixes and possessor raising that add
one, but unlike some Bantu languages such as KiHaya (Duranti and Byar-
ushengo 1977), in no case can a verb be derived that exceeds the number of
arguments of an underived verb; the maximum number of core arguments
of any verb is three. Beyond these processes of decreased and increased
transitivity, there are no derivational processes in Yimas that affect a verb
and its arguments’ mapping into constituent structure. There are no voice
alternations such as passive or antipassive and no alternations in realization
such as we find in English between the farmer loaded hay on the truck versus the
farmer loaded the truck with hay. Given a verb’s logical structure and the
arguments it subcategorizes, there is one and only one mapping into con-
stituent structure that corresponds to it. The case marking of the arguments
is very complex and will be discussed in the following section; here I am just
concerned with their basic realization as constituents. Consider an intransi-
tive unaccusative verb like mal- ‘die’, as in na-mal 3sg.nom-die ‘he just died’.
This is an achievement predicate and so has the following logical structure:
INGR died′ (he). As he is an argument in a frame of pred′ (x), it will be
assigned the undergoer macrorole by the universal Actor-Undergoer Hier-
archy (AUH), and as the sole argument will appear in nominative case as a
bound prefix to the verb. Now consider an intransitive unergative verb like
iray ‘cry’ as in na-iray 3sg.nom-cry ‘she cried’. This is an activity predicate
with the logical structure do′ (she, [cry′ (she)]). As she is an argument in a
frame of do′ (x), by the universal AUH it will be assigned the actor macro-
role, but again as the sole argument will appear in nominative case as a
bound prefix to the verb. Now consider a canonical transitive verb such as
warapa- ‘cut with a flat sharp instrument’ as in (5).
(6) [do′ (I, Ø)] CAUSE [[ do′ (shell, Ø)] CAUSE [INGR cut′ (it)]]
By the universal AUH, it as the argument of pred′ (x) would be assigned the
undergoer macrorole and therefore be realized by nominative case in this
direct arrangement of arguments (see Section 23.5). But there are two
potential actors, I and shell, as both are first arguments of do′ (x, . . .) (Yimas,
unlike Tagalog, has no constructions which differentiate grammatically
between volitional deliberate actions and those which are not, and there
are no verb roots which lexicalize this contrast like see versus watch, so there
is no need to postulate an operator DO in logical structures.) Yimas, unlike
English, does not permit sentences like the shell cut it; the animate instigator
who instigates the causal chain must be mentioned overtly and outranks the
instrument. Hence it is the animate instigator, here I, which is assigned the
actor macrorole and receives ergative case, realized in the core along with
the undergoer, while the instrument is realized as an oblique RX in the
periphery of the core. Finally, consider the canonical ditransitive verb ŋa-
‘give’ as in (7).
SG mn n- na- -nakn
DL mrm mp- impa- -mpn
PC mŋkt ŋkl- kra- -ŋkt
PL mum mpu- pu- -mpun
these affixes, two principles apply, both aspects of the direct-inverse system
of the language: first, local persons outrank non-local persons, specifically
1st > 2nd > 3rd, and second, surprisingly, accusative prefixes outrank
ergative ones which in turn outrank nominative ones. There are also two
purely structural constraints: one, the most highly ranked prefix must
appear in the immediately preverbal position, and, two, every verb should
have a nominative affix, and that must occur on the left edge of the verb in
initial position, unless usurped or demoted by a modal or other inflection to
the right edge, as in (4). Let’s see how this system all works. Consider how
one would say ‘I hit them’ versus ‘they hit me’ in Yimas. The direct form, ‘I
hit them’ (local person acting on non-local person) is straightforward. The
local person is the transitive actor and ergative, hence ka- 1sg erg. The non-
local person is the transitive undergoer, and the case-marking system for the
non-local person is ergative-nominative, so the undergoer is realized as
nominative, pu- 3pl nom, simultaneously satisfying the requirement for
an overt nominative. Because local > non-local and ERG > NOM, the first-
person ergative prefix will occupy the salient immediately preverbal pos-
ition, and the third-person nominative the left edge.
(9) pu-ka-tpul
3pl.nom-1sg.erg-hit
‘I hit them.’
The inverse form ‘they hit me’ is a little more complicated. The undergoer
is a local person, and local persons have accusative forms (e.g. ŋa- 1sg acc).
Note that the local person is higher by both local > non-local, and ACC >
ERG, so ŋa- 1sg acc must occupy the salient immediately preverbal position.
The non-local actor would normally take ergative case, and indeed there is
an ergative form for third plural: mpu- 3pl erg. But the expected form *mpu-
ŋa-tpul is ungrammatical because it runs afoul of the requirement for an
overt nominative; such a verb lacks a nominative prefix on the left edge. So
instead, the third plural pronominal is realized by the corresponding nom-
inative prefix, pu- 3pl nom.
(10) pu-ŋa-tpul
3pl.nom-1sg.acc-hit
‘They hit me.’
(11) k-ka-ŋa-r-mpun
vi.sg.nom-1sg.erg-give-perf-3pl.dat
‘I gave it to them.’
(12) k-mpu-ŋa-ŋa-t
vi.sg.nom-3pl.erg-1sg.acc-give-perf
‘They gave it to me.’
Let’s see how we could handle this very complex case system by extending
to Yimas case assignment the rules for ergative/accusative constructions
provided in Van Valin (2005: 108), reproduced here as (13). (There is a much
more detailed and typologically varied description of these in Van Valin and
La Polla (1997), but the points made below would still apply.)
It is clear that these will need some modification for Yimas. First, we
have no need for an absolutive case, and invoking it would greatly
complicate the description. So-called absolutive case is simply nomina-
tive, in Yimas and, I would argue, universally. That is why, like nomina-
tive in accusative constructions, erstwhile absolutive case constituents
are normally formally unmarked and why the nominative is always the
privileged case in Yimas regardless of whether it shows up in the
accusatively aligned local persons or the ergatively aligned non-local
persons. Further, the problem of the lower-ranked macrorole in (13a) is
problematic in the case of ditransitive verbs with two undergoers, but
let us see how we go. For non-local persons the modification is very
straightforward: simply assign nominative case to the lowest argument
on the AUH (and dative case to the other one if present) and ergative
case to the highest macrorole. The rules for case assignment to the local
persons are more complex, because the rules for accusative and ergative
case constructions apply simultaneously and partially, that is, only
something like the (b) conditions apply. If we were to apply all the
conditions, we will get case conflict, a transitive actor will be assigned
nominative case by accusative condition (a), but ergative by ergative
condition (b). But if we absorb the (a) conditions into and restate the
(b) conditions, the system will work: (a) assign accusative case to the
lower-ranked macrorole when there are at least two (there is a compli-
cation with ditransitive verbs with local recipients; there assign accusa-
tive case with the next to highest (or lowest, either will work) argument
on the AUH); (b) assign ergative case to the highest-ranked macrorole,
again where there are at least two; (c) assign nominative to any core
argument that has not received case by (a) or (b). The tree diagram in
Figure 23.3 represents the mapping of (11) using these rules from logical
structure to constituent structure.
SENTENCE
CLAUSE
CORE
RX RX NUCLEUS RX
PRED
In (14a) the argument shared between the clauses and elided from the first
clause is the undergoer, while in (14b) there are no shared arguments. Some
In (15a) the actor of a transitive verb papk ‘carve’ has been genitivized in the
nominalized non-finite complement, but if we try this with the undergoer in
(15b), the result is ungrammatical (though okay in English). However, with
the single argument of an intransitive verb (15c), genitivization is again
possible. Other constructions, though, work on an intransitive subject/tran-
sitive object (S/O) pivot, for example the scope of elevational/directional
affixes. Consider these examples.
(16) a. kay i-ɲa-l-ampu-n
canoe.viii.sg viii.sg.nom-pres-down-float-pres
‘The canoe is down there.’
b. kay naŋ-l-arm-na-ŋkan-i
canoe.viii.sg imp.pl-down-board-imp-pc-viii.sg.nom
‘You all board the canoe down below.’
‘*You all down below board the canoe.’
Note that in (16a) the scope of the elevational prefix l- is over the sole
argument of the intransitive verb ampu- ‘float’, while in (16b) it is over the
undergoer of the transitive verb arm- ‘board’ and cannot be construed as
modifying the actor, a clear S/O pivot. The only way to say the equivalent of
the starred translation would be something like ‘you all are standing/sitting
down below and now you all board the canoe’.
RX
COREN
CORE NUCLEUSN
RX NUCLEUS RX
PRED
V N
NUCLEUSN
COREN NUM
RX
These are marked by specific noun class marker suffixes utilized only for
this function, or in the case of complements of desire by a bound noun
wampuŋ, literally ‘heart’. Furthermore, as true complements they function
as core arguments of the complement-taking verb of the core of the main
clause and are therefore commonly (17a, b, c), though not obligatorily
(17d), indicated there by a corresponding bound verbal affix of the same
noun class.
b. complement of activity
nam wark-t-nti tia-ka-ira-karŋkra-t
house enclose-nfn-act act.nom-1SG.erg-appl-tired-perf
‘I’m tired of building houses.’
c. complement of customary or habitual action
yaki am-t-wal ntak-na-k
tobacco.v.pl eat-nfn-custom.v.sg leave-imp-V.sg.nom
‘Stop smoking!’
d. complement of desire
tpuk am-t-wampuŋ kpa-n ama-na-t-n
sago.x.sg eat-nfn-heart.v.sg big-v.sg 1sg.nom-pres-feel-pres
‘I really want to eat sago.’ (‘I have a big desire to eat sago’)
Note in (17d) that the adjective kpa ‘big’ modifies the nominalization.
Core arguments of non-finite nominalizations may be the pivot of control
constructions and when they are, the pivot is strictly S/A; undergoers may
never be controlled. Otherwise, genitivization applies, again strictly under
an S/A pivot, illustrated above in (15) and in this example.
The controllers of the S/A pivot in control constructions can be any of the
three core arguments, and what determines which is the Theory of
Obligatory Control (Foley and Van Valin 1984: 307–311; Van Valin 2005:
243): (1) causative and jussive verbs have undergoer control and (2) all
other transitive verbs have actor control. In control structures in Yimas
the nominalized complement always functions as an undergoer, so with
simple transitive complements there is no issue: the controller can only
be the actor, the only other macrorole, as in (17b, d) above and in the
examples below.
Ditransitive verbs are more complicated as there are two potential con-
trollers, the actor or the second undergoer. In accord with the Theory of
Obligatory Control, if the complement is a causative or jussive one, there is
undergoer control, as in (17a) and (20).
In (24a, b) the actor of the nominalizations is the same as that of the matrix
verb, and being such, it can be elided. But such control is not obligatory, as
(24c) demonstrates. There the actor of the nominalization is num-un-mat
village-obl-i.pl ‘villagers’, while that of the matrix verb is arm ‘water’, which
is always formally plural in Yimas, hence the bound prefix mpu- 3pl.erg. Any
core argument can be the controller of an elided actor of an oblique non-
finite nominalization.
kalc-r-awt kalc-k-n
strengthen-nfn-i.sg strengthen-irr-i.sg
‘hard worker’ ‘a strong person’
am-t-awt am-t-ra
eat-nfn-i.sg eat-perf-v.pl
‘eater’ ‘food’
Most relative clauses are rather more elaborate than this. The verb of the
relative clause (and many are just the verb) is typically suffixed as a whole to
the near distal deictic stem m- ‘that’. Further, they can have bound
RX
COREN
CLAUSE NUCLEUSN
CORE
NUCLEUS RX
PRED
V N
wa-kia-k -ra
go-NEAR-IRR V.PL
NUCLEUS NUCLEUSN
TENSE CLAUSE RX
Example (27a) illustrates a case where the head and relativized argument is
the intransitive subject. This single core argument is realized solely by the
head suffix, the verb bearing no other bound pronominal bound affix. In
(27b) the relativized noun and head is the actor of a transitive verb, realized
by the final head suffix, while the undergoer is realized by its proper prefix.
Finally, in (27c) the relativized noun and referent of the final head suffix
corresponds to the undergoer argument of a transitive verb, while the actor
appears now as a prefix in its proper position.
All types of core arguments and even peripheral adjuncts are relativizable
(though recipient undergoers with difficulty due to morphological compli-
cations (see Foley 1991: 417–418)); here are some examples.
(28) a. on actor
Elias [m-kra-pay-pra-kia-ntuk-ŋkt-ø mota-nan]
pn nr.dist-1pl.acc-carry-toward-near-rm.past-pc-i.sg motor-obl
‘(It was) Elias who brought us few by motor.’
b. on undergoer
anti [God m-n-papk-ntuk-i]
ground.viii.sg God nr.dist-3sg.erg-carve-rm.past-viii.sg
‘the world which God made’
panmal m-n-tpul-c-(n)ak narmaŋ
man.i.sg nr.dist-3sg.erg-hit-perf-i.sg woman.ii.sg
‘the man which the woman hit’
c. on theme undergoer
impram [m-(n)-nan-(n)a-ampa-ŋa-nt-m]
basket.vii.sg nr.dist-3sg.erg-2pl.acc-pres-weave-appl-pres-vii.sg
‘the basket which she is weaving for you’
d. on peripheral location (again the fossilized suffix -a ix.sg is employed)
[maramara m-mpu-t-r-a-n]
goods.v.pl nr.dist-3pl.erg-lay.down-perf-ix.sg-obl
m-ra ya-kay-ɲa-tacay-kulanaŋ
nr.dist-v.pl v.pl.nom-1pl.erg-pres-see (red: tay-)-walk
‘Where they laid out the goods, we walked around looking at them.’
(Literally ‘at the place that they laid out the goods. . .’)
e. as possessor
[awt m-nanaŋ-pampay-caw-na-ntut- ø] na
fire nr.dist-dur-appl-sit-dur-rm.past-i.sg poss
kalakn na-n-tay-mpi-yara-k
child.i.sg 3sg.nom-3sg.erg-see-seq-pick.up-irr
‘The son of him who was tending the fire found it.’
f. as possessed
manm p-ka-tay m-ɲa
cult.house.vii.sg vii.sg.nom-1sg.erg-see nr.dist-poss
[m-mpu-tkam-r-m]
nr.dist-3pl.erg-show-perf-vii.sg
‘I saw the cult house, yours, which they showed.’
Figure 23.6 illustrates the proposed representation for the relative clause of
the second example in (28b).
What is unusual about Yimas in comparison to more familiar languages is
that the entire relative clause, with the exception of the ‘leaked’ core
argument narmaŋ ‘woman’, corresponds to a single word, and such words
can be quite complex (28a, e). But that is entirely in keeping with the
polysynthetic typological profile of the language; the word is its favoured
grammatical unit and structures above that are very weakly syntactically
developed. This fact should again caution us against building grammatical
theories solely on the basis of what is familiar to us, such as constituency:
they blind us to other possibilities.
The final type of relative clause to consider are those in which the relativ-
ized noun is a temporal expression. These are obliquely case-marked like
relative clauses on locative adjuncts, and, also like them, the head of these
temporal relative clauses is a fossilized noun class and number suffix, of the
form -mp vii.sg. But in this case we know the noun to which it refers
because it is still synchronically in the language: pucm ‘part, piece, time’.
All the semantic relations on the Interclausal Semantic Relations Hierarchy
that are expressed typically by adverbial subordinate clauses in more famil-
iar languages, circumstances, reasons, conditionals and hypotheticals and
concessives, are expressed by this type of temporal relative clause in Yimas.
The differences between them are captured in the choice of tense and mood
for both the relative clause and the matrix clause. For example, for temporal
circumstances, the verb of the relative clause occurs with one of the past or
present tenses or that of the definite future.
(29) [m-mpu-ŋa-na-tay-ɲc-mp-n]
nr.dist-3pl.erg-1sg.acc-pres-see-pres-vii.sg-obl
pu-ka-apan-kt
3pl.nom-1sg.acc-spear-rm.fut
‘When they see me, I will spear them.’
But conditionals and hypotheticals are marked with mood affixes that
mark unreal or as yet unrealized events.
RX
COREN
CLAUSE NUCLEUSN
CORE
NUCLEUS
PRED
RX V RX N RX
m- n- tpul-c-
t l-c-
tpu -(n)ak
-(n
(n)ak narma
narm
r
NR.DIST 3SG.ERG hit-PERF I.SG wom
woman
NUCLEUS NUCLEUSN
TENSE CLAUSE
EIC
DEIC RX
23.8 Conclusion
References
Abbreviations
24.1 Introduction
seen in example (3), but they are interpreted like activities in progress with
the progressive (e.g. ‘the door is opening’).
Semelfactives are events that lack duration. Like activities, they are inter-
preted as past actions in the perfective (e.g. ‘she came’). However, unlike
activities, they cannot be modified by duration adverbials such as ‘for 5
minutes’ and ‘quickly’. They can be used with the progressive, in which case
they are interpreted as iterative or repeated actions, as in example (4).
As yet, no clear diagnostic for the classes achievement and active accom-
plishment has been found for Avatime. The typical test using the modifiers
‘for X time’ or ‘in X time’ is not applicable as this specific semantic distinc-
tion is not consistently made. There is, however, a group of atomic telic
predicates where the endpoint is reached in a single step rather than
gradually (Caudal and Nicolas 2005), which can be distinguished based on
their incompatibility with the adverb petee ‘completely’, as in example (5).
(6) -d ̀ bàsi¼ɛ
c c a-dra ò-pópo¼lò
_
c1.sg-teacher¼def c1.sg.sbj.pfv-open c2.sg-door¼def
‘The teacher opened the door.’
Logical Structure: do′ ( ̀ d ̀ bàsi, Ø) CAUSE [BECOME open′(òpópo)]
c c
Sentence
Clause
RP Core Periphery
Pro Nuc RP
Pred
V PP
ɔd
́ zɛɛ á- ta áʋanà ní lịgbalɛ mɛ̀
(‘direct/indirect object’) follow the verb. Adverbial phrases follow the post-
verbal arguments. There is head-marking of the PSA on the verb. A canonical
transitive sentence with an adverbial phrase in Avatime can be seen in
example (7) and the syntactic structure of this example is shown in
Figure 24.1. Following Van Valin (1985) and Van Valin and LaPolla (1997),
we analyse the PSA-indexing prefix on the verb as an argument in the core,
whereas the RP with which this prefix is coreferential is analysed as occur-
ring outside the core, but inside the clause. This captures the observation
that a lexical RP or independent pronoun is not necessary, and example (7)
remains grammatical when leaving out the RP ́ dzɛɛ ‘the woman’. c
Avatime makes use of the pre-core slot (PrCS) for question formation and
focus marking (see also Section 24.7 and Van Putten 2016). The question
word or focused RP is placed in the pre-core slot and marked with a final
extra-high tone, as shown in examples (8) and (9).
egé ́ -dzɛ¼ɛ a-ta
c
(8)
what c1.sg-woman¼def c1.sg.sbj.pfv-chew
‘What did the woman eat?’ (elic-QUIS-foc_100714_SO)
Sentence
PrDP Clause
PrCS RP Core
Pro Nuc
Pred
RP RP V
kivoe áʋaná ɔd
́ zɛɛ á- ta
Figure 24.2 A sentence with pre-detached position and pre-core slot (example 11)
24.4.2 Operators
In this section, we give an overview of the operators that have scope over the
different layers of the layered structure of the clause. The Avatime verb is
usually marked with a prefix that indexes the PSA. This prefix (henceforth
referred to as the PSA prefix) also encodes aspect or mood. It is optionally
followed by other aspect, mood and directionality prefixes. There is no tense
marking. An overview of the categories marked on the verb and the order in
which they occur can be seen in (12) and an example with all slots occupied
can be seen in (13).
Set 2
- Negative (with extra-high Set 3
tone) - Subjunctive
Set 1 - Subjunctive (some verbs) (some verbs)
Person/ number/ Perfective - Perfective (locative and - Habitual (with
noun class (most verbs) copula verbs) zɛ̌ - prefix) Potential Progressive
There are five paradigms of PSA prefixes, as shown in Table 24.1. For the
sake of brevity, the forms for the noun classes other than class 1 singular
and plural have been omitted (a full table can be seen in Van Putten 2014a or
Defina 2016a). As the table shows, there is no one-to-one correspondence
between each paradigm and an aspect/mood category. It is also clear that the
prefixes are formally related to each other; it is likely that they have evolved
from combinations of a prefix indexing the PSA plus an aspect/mood
marker, but they cannot be synchronically analysed as such.
We will now briefly discuss the operators one by one. For more detail
about these markers and their semantics, see Defina (2018).
Given that aspect marking modifies the internal temporal structure of the
event itself, the aspect markers are considered to be nuclear-level operators
(Van Valin and LaPolla 1997; Van Valin 2005). Avatime has four aspectual
categories: perfective, progressive, habitual and recurrent. Their forms are
described in (14).
There are two directionals: the itive (‘away’), marked with the prefix ze-/zɛ-
and the ventive (‘towards’), marked with the prefix bá-/bé-, shown in example
(17). These forms express orientation or motion away from or towards the
deictic centre. This means they are core-level operators (see Van Valin and
LaPolla 1997: 45).
(19) mla-trɛ
2.pl.sbj.sbjv-go
‘You (pl) should go.’ / ‘Go! (to multiple people)’
particle na or, more frequently, with intonation only. Content questions are
marked with a question word in the pre-core slot (see example (8) in Section
24.4.1).
According to RRG (Van Valin and LaPolla 1997; Van Valin 2005), the
order of operators in a given language always follows the same pattern
whereby clausal operators occur furthest from the verb, core operators
closer than clausal operators, and nuclear operators are the closest to the
verb. In Avatime, this prediction is mostly borne out, but there are two
exceptions: (i) the (core) directional markers occur closer to the verb than
the (nuclear) aspect markers, and (ii) the (core) intentive marker also
occurs closer to the verb than the (nuclear) perfective marker. The direc-
tional markers have clearly grammaticalized from motion verbs (ba ‘come’
and za ‘pass’). Before grammaticalization, they would have occurred as the
first verb in a serial verb construction (SVC), directly followed by the
second verb (see Sections 24.9 and 24.10 for more information about serial
verb constructions). As all aspect and mood marking in SVCs is on the first
verb, it now precedes these directional prefixes. A similar explanation
could account for the intentive following the perfective: the intentive tá-
is also likely to be the result of a grammaticalization process from a serial
verb construction, probably from the verb trɛ ‘go’ or tráà ‘come’. Another
possible explanation for the intentive following the perfective is the
hypothesis that the perfective forms used to be the bare subject agreement
markers unmarked for aspect/mood, as is the case in many related lan-
guages (Welmers 1973). Through language change, the default PSA prefix
came to be used solely for perfective marking and thereby turned into a
nuclear-level operator, which happened to occur in a position preceding
the intentive marker.
Figure 24.3 shows the sentence in example (13), repeated here as (20), with
both constituent and operator projections.
Sentence
Clause
Core
NP Nuc NP
Pred
V
V
ASP
Nuc
ASP Nuc
DIR Core
MOD Core
STAT Clause
Figure 24.3 Sentence from example (20) with constituent and operator projections
carries a noun-class prefix which indicates its gender and number. Within
the referential phrase, there is agreement on numerals, articles and demon-
stratives (see example (24)). For more detailed information on the noun-class
system and the various nominal modifiers, see Schuh (1995), Van Putten
(2014a) and Defina (2016a).
The constituent order in Avatime RPs is shown in (21). Adjectives are
considered to occur in a periphery of the nucleus, following Van Valin
(2005), whereas the other nominal modifiers are operators on different
layers of the RP.
The only operator on the nuclear level is number, which is marked on the
noun class prefix (e.g. ̀ -kli ‘leg’ vs. ı̀ ̣-kli ‘legs’). The numerals are operators on
c
_ _
the core level, as they provide quantification. The determiners are oper-
ators on the RP level, as they ground the RP in discourse. The class of
determiners contains the elements shown in (22), of which only one can
occur in a given RP.
(22) Determiners
• The definite article, which is a monosyllabic enclitic and agrees with the
noun class of the head noun.
The class of particles contains words such as tsyɛ ‘also’, k ‘by contrast’ and c
kò ‘only’. Like definiteness and deixis markers, they ground the RP in
discourse and are therefore considered to be RP-level operators.
Possession in Avatime is mostly indicated by juxtaposition with a possessor–
possessed word order. The regular independent pronouns are used as posses-
sor pronouns. For possessed kinship terms, there is a slightly more complex
construction in which the possessor pronoun fuses with the noun-class prefix
of the possessed kinship term.3 With a nominal possessor, this fused possessor
pronoun is also present, and, thus, in these cases there is head-marking of the
possessor on the possessed noun. Examples of kinship and non-kinship terms
with nominal and pronominal possessors can be seen in (23).
(23) yɛ ‘c1.sg’ þ c
̀ -mà-n ‘the town’
c ! yɛ ̀ -mà-n ‘her town’
c c
yɛ ‘c1.sg’ þ o-ne ‘mother’ ! ye-ne ‘her mother’
c-nùv ̀ -ɛ ‘the child’
c þ li-kuto-lɛ ‘the hat’ ! nùv ̀ -ɛ li-kuto-lɛ ‘the child’s hat’
c c
c-nùv ̀ -ɛ ‘the child’
c þ o-ne ‘mother’ ! nùv ̀ -ɛ ye-ne ‘the child’s mother’
c c
The possessor occurs in the RP initial position (RPIP), analogous to both the
pre-detached position and pre-core slot in the layered structure of the clause
(see Van Valin 2005: 26).
An example of a complex RP in Avatime can be seen in (24). The constitu-
ent and operator projections of this example are shown in Figure 24.4.
RP
RPIP CoreR
RP
CoreR
NucR R NucR
N MP N
N N
NUM NucR NUM NucR
CoreR QUANT CoreR
RP DEF RP DEF
Figure 24.4 Complex RP with constituent and operator projections (example 24)
PP
CoreP
RP
RPIP CoreR
NucP RP NucR
Pred
PREP POSTP
ní ɔk̀plɔnɔ ̀ abà
with the locatum in a possessive construction (e.g. ‘the upper surface of the
table’ ! ‘on the table’, see also Section 24.2).
In locative phrases, a postpositional phrase must be combined with the
preposition ní, as shown in example (25). The constituent projection of this
ACTOR UNDERGOER
----------------------------------------------------->
<- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
After assigning macrorole status, the PSA is selected. In Avatime, the PSA is
the argument that is indexed on the verb. Avatime has a nominative-
accusative alignment pattern: if there is only one macrorole argument, this
is selected as the PSA, and if there are two macroroles, the actor is selected as
the PSA. The PSA is invariable: there is no passive construction in Avatime
and non-macrorole PSAs are not allowed. In example (28), the actor, Kofi, is
selected as the PSA.
After the selection of the PSA, the appropriate syntactic template is
selected. The syntactic template for a canonical Avatime sentence has one
core slot before the verb for a head-marked argument and up to two core
slots for RP arguments after the verb (see Section 24.4.1). In the case of
example (28), a syntactic template with two core slots after the verb is
chosen, as there are three arguments in the logical structure. The preverbal
head-marking slot is linked to the PSA, which is Kofi. This slot can only
contain PSA-indexing prefixes and does not allow referential phrases. There-
fore, it is filled with a PSA-indexing prefix of noun class 1 singular (class 1 is
used for people) and the RP Kofi is placed in the pre-core position (see also
Section 24.4.1). The undergoer and non-macrorole arguments are linked to
the two postverbal core-internal slots.
marked with the final extra-high tone, while the inflected verb remains in
its normal position. An example is shown in (30), where the focus is on the
lexical content of the verb, but the same construction can also be used
to mark focus on an aspect, modality or status operator (see also Van
Putten 2016).
(30) ki-h ́
c bɛ-tá-h c lc
c4.sg-grind:foc c1.pl.sbj.pfv-int-grind c3.sg
‘Will they [grind]FOC it (or pound it)?’ (illness_100616_SO-DS)
The focus construction may also be used for sentence focus or predicate
focus (focus on the verb þ argument that is not the PSA or PP). A sentence
focus interpretation is possible when the PSA is marked for focus and a
predicate focus interpretation is possible when an argument that is not the
PSA or a peripheral PP is marked for focus. An example of the latter can be
seen in (31).
(31) (One speaker is telling a story in which a boy riding a bicycle meets a girl. The
listener asks a clarification question.)
yɛ tsyɛ, gas ́ c ὲέ-kpɛ
c1.sg too bicycle:foc c1.sg.sbj.prog-put
‘Is she also [riding a bicycle]FOC ?’ (pear_100630_GoD-FB)
For the linking from semantics to syntax, the use of the focus construction
means that a syntactic template with a pre-core slot will be chosen (see also
Section 24.4.1). If the element to be placed in the pre-core slot is a non-PSA
argument, the number of postverbal core slots will be reduced by one. PSA
indexing on the verb is obligatory, so the preverbal core slot which contains
the verb prefix cross-referencing the nominal PSA argument is always filled,
irrespective of whether the PSA occurs in the pre-core slot or in its regular
pre-core position.
24.8 Summary
Serial verb constructions are sequences of two or more finite verbs in one
clause without overt marking for coordination or subordination. In Ava-
time, only the first verb is fully inflected. Subsequent verbs are either bare,
as in example (32) or marked with a reduced agreement prefix, as in
example (33). These reduced prefixes are specific to SVCs (for more infor-
mation, see Defina 2016b). All verbs share their PSA and may also share
another argument, as is the case with example (33).
Avatime SVCs can be divided into three subtypes: nuclear, core and
sequential, based on a number of morphosyntactic differences. In what
follows we briefly describe the properties of the different types of SVC and
show how they can be analysed in RRG. For a more elaborate account, see
Defina (2016b).
The different subtypes can be distinguished by examining their behav-
iour with operators at different levels. Negation and most aspect and mood
markers are fused to some degree with the PSA agreement markers (see
Section 24.4.2) and can only be marked on the first verb of an SVC. These
markers always have broad scope over the whole SVC. Even the intentive,
which is the only morphologically separate mood marker, can only occur
with the first verb and scopes over the whole construction. This suggests
that the verbs in all Avatime SVCs are linked below the clause level.
Aspect and aspectual adverbials are nuclear-level operators and can be
used to distinguish a set of nuclear-level SVCs from the rest. Many SVCs
allow aspectual adverbials between the verbs with narrow scope over a
single verb, as can be seen in example (34). The recurrent aspect can also
be used to mark each verb in these SVCs individually, as can be seen in
example (35). Neither of these is possible with the nuclear-level SVCs. While
aspectual adverbials are occasionally accepted between the verbs of these
nuclear SVCs, they always have broad scope over the whole construction and
the recurrent aspect can only be marked on the first verb, as can be seen in
examples (36) and (37). This indicates that these SVCs are formed through
nuclear co-subordination while the other Avatime SVCs are joined at a
higher level.
The other non-nuclear SVCs also allow locational and temporal adverbials
between the verbs. However, in these cases there can be only one adverbial
and it has broad scope over the whole construction, as in example (40). This
suggests these SVCs share a single periphery and are joined via core co-
subordination. Nuclear SVCs do not allow locational or temporal adverbials
to occur between the verbs, consistent with their nuclear juncture.
Strongest Closest
Nuclear co-subordination Causative [1]
Phase
Nuclear subordination Manner
Daughter Motion
Peripheral Position
Nuclear coordination Means
Psych-action
Core co-subordination Purposive
Jussive
Core subordination Causative [2]
Daughter Direct perception
Peripheral Indirect perception
Core coordination Propositional attitude
Cognition
Clausal co-subordination Indirect discourse
Direct discourse
Clausal subordination Circumstances
Daughter Reason
Peripheral Conditional
Clausal coordination Concessive
Simultaneous actions
Sentential subordination Sequential actions
Situation-situation: unspecified
Sentential coordination
Weakest Loosest
Figure 24.6 The Interclausal Relations Hierarchy (Van Valin 2005: 209)
from the hierarchy. It is also notable that there is a semantic split between
the functions of the directional prefix and the motion nuclear SVCs where
this deviation occurs. It could then be possible to split motion into two
functions with accompanying motion positioned as closer than manner,
and path of motion remaining where motion is currently (looser than
manner). This would resolve the deviation in the hierarchy.
The core SVCs are used for adding and introducing or marking arguments.
Examples (34) and (40) show the use of this type of SVC to add a beneficiary
using the verb kí ‘give’. They can also be used to add instruments, manners
or means, as in examples (43)–(45), respectively, all with the verb k ̀ ‘take’. c
(43) a-k ̀
c kà-wɛ¼a yài ò-se¼lò
c1.sg.sbj-take c6.sg-axe¼def break c2.sg-tree¼def
‘He used an axe to split the tree.’
(44) a-k ̀
c ku-siyeyome sὲ
c1.sg.sbj.pfv-take c5.sg-anger leave
‘He left in anger.’
(45) a-k ̀
c ku-zò dzi ohonete
c1.sg.sbj.pfv-take c5.sg-theft become rich.person
‘Through theft he became a rich man.’
These SVCs can also be used to mark or introduce theme arguments which
are already called for by another verb in the SVC. For instance, in example
(46) k ̀ ‘take’ is used to introduce the theme banana. However, as example
c
(47) shows, kpɛ ‘put’ already licenses agent and theme arguments.
(46) a-k ̀
c k ̀ ranti¼ɛ
c kpɛ ní kà-s ¼ya
c mὲ
c1.sg.pfv-take banana¼def put loc c6.sg-basket¼def inside
‘S/he put the banana into the basket.’
With the exceptions of manner and means, these functions are not
included within the list of interclausal semantic functions. These
argument-adding functions are relations between a predicate and its
dependents and are typically intraclausal. However, the fact that they
can be expressed by a combination of verbs in Avatime suggests they
could be added to the hierarchy. They bear clear similarities to the
causative [1] function – to directly bring about a state of affairs through
an event or action: for example, paint the table red or break the bowl. This
function also involves relations between a predicate and its arguments
and is sometimes expressed by a single verbal lexeme. Given the overlap
with manner and means functions and the general role of these func-
tions, it would make sense for them to be added immediately below
means in the hierarchy.
24.11 Summary
This part has surveyed the semantic functions and syntactic properties of
SVCs in Avatime, showing how they constitute three distinct juncture–
nexus types, see Table 24.3.
Each subtype of Avatime SVC poses a question for the Interclausal
Relations Hierarchy and its predicted links between syntactic
References
Defina, Rebecca. 2016a. Events in Languages and Thought: The Case of Serial Verb
Constructions in Avatime. PhD dissertation, Radboud University, Nijmegen.
Defina, Rebecca. 2016b. Serial verb constructions and their subtypes in
Avatime. Studies in Language 40(3): 648–680.
Defina, Rebecca. 2018. Tense, aspect and mood in Avatime. Afrika und Übersee
92: 65–97.
Dingemanse, Mark. 2012. Advances in the cross-linguistic study of ideo-
phones. Language and Linguistics Compass 6(10): 654–672.
Foley, William A. and Robert D. Van Valin, Jr. 1984. Functional Syntax and
Universal Grammar. Cambridge: Cambridge University Press.
Givón, T. 1980. The binding hierarchy and the typology of complements,
Studies in Language 4(3): 333–377.
Kockelman, Paul. 2003. The Interclausal Relations Hierarchy in Q’ecqchi’
Maya. International Journal of American Linguistics 69: 25–48.
van Putten, Saskia. 2014a. Information Structure in Avatime. PhD dissertation,
Radboud University, Nijmegen.
van Putten, Saskia. 2014b. Left-dislocation and subordination in Avatime
(Kwa). In Rik van Gijn, Jeremy Hammond, Dejan Matić, Saskia van Putten
and Ana Vilacy Galucio (eds.), Information Structure and Reference Tracking in
Complex Sentences, 71–98. Amsterdam: John Benjamins.
van Putten, Saskia. 2016. Discourse functions of focus marking in Avatime.
Journal of African Languages and Linguistics 37(1): 91–130.
Schuh, Russell G. 1995. Avatime noun classes and concord. Studies in African
Linguistics 24(2): 123–149.
Silverstein, Michael. 1976. Hierarchy of features and ergativity. In R. M. W.
Dixon, ed., Grammatical Categories in Australian Languages, 112–171. Can-
berra: Australian Institute of Aboriginal Studies.
Van Valin, Robert D., Jr. 1985. Case marking and the structure of the Lakhota
clause. In Johanna Nichols and Anthony C. Woodbury (eds.), Grammar
Inside and Outside the Clause, 363–413. Cambridge: Cambridge University
Press.
Van Valin, Robert D., Jr. 2005. Exploring the Syntax-Semantic Interface. Cam-
bridge: Cambridge University Press.
Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax: Structure, Meaning,
and Function. Cambridge: Cambridge University Press.
Welmers, William E. 1973. African Language Structures. Berkeley, CA: Univer-
sity of California Press.
Notes
1 All recordings are archived at The Language Archive at the MPI for
Psycholinguistics in Nijmegen (https://corpus1.mpi.nl/ds/asv/?1). Examples
in this chapter that are taken from this corpus are followed by the
filename of the recording.
2 This prefix is different from the habitual prefix, despite being homoph-
onous with it. See Defina (2018) for more details.
3 We could call this inalienable possession, but unlike most languages that
have an inalienable vs. alienable distinction, this class consists only of
kinship terms and does not include body parts.
Abbreviations
25.1 Introduction
* Robert Van Valin provided helpful comments and suggestions in the compilation of this chapter. However, any errors in
analysis or content are the responsibility of the author.
There are two basic types of M-intransitive motion verb: those that specify
[moveþpath] and those that specify [moveþmanner]. [moveþpath]
motion verbs are verb-framed and [moveþmanner] motion verbs are
satellite-framed (Talmy 2007). Each type of motion verb is M-intransitive
and the single argument is actor in each case. However, there is a syntactic
difference between these types of motion verb. The [moveþpath] motion
Other verbs that are M-intransitive include ededec ‘gleam’ do′ (x, [gleam′ (x)]),
asalec ‘laugh’ do′ (x, [laugh′ (x)]), busuec ‘fart’ seml fart′ (x), silolec ‘ooze’ do′ (x,
[ooze′ (x)]), fojec ‘vomit’ do′ (x, [vomit′ (x)]), tatiec ‘look up’ do′ (x, [look.up′ (x)]),
bodoec ‘soften’ become soft′ (x), sanan mec ‘start’ ingr start′ (x).
Impersonal verbs also have to be treated as M-intransitive. A typical
example of an impersonal verb construction (IVC) is given in (6). There is
an optional free pronoun ija ‘1sg’ which corresponds to the subject pronoun
in the English translation. However, in the Amele form this pronoun is cross-
referenced on the verb with acc morphology. The nom agreement always
codes 3sg in an IVC but the reference is unspecified. The agreement is
dummy, neutral agreement which is part of the syntactic template for the
construction and there is no actor argument with these IVCs. Such IVCs
normally express a physiological or psychological experience and the 1sg.
acc agreement refers to the experiencer argument.
know′ (x, (y)), gawec ‘want’ want′ (x, y), meleec ‘believe’ believe′ (x, (y)), cucuiec
‘fear’ fear′ (x, (y)). With some M-transitive non-state verbs, expression of the
undergoer argument on the verb is optional, with others this is obligatory.
(7) illustrates an active verb, j-ec [eat-inf] ‘to eat’, with optional DUn marking
on the verb. In (7a) this verb has the logical structure (LS) do′ (x, [eat′ (x, (y))])
with an optional second predicate argument. In (7b) this verb has the LS do′
(dana, [eat′ (dana, ha)]) with the second predicate argument position filled.
This argument is realized in the syntax by the RP ha ‘sugarcane’. Since ha is
neither animate nor a count noun, in this context it does not trigger acc
agreement on the verb. The predicate is, therefore, an activity with a non-
specific object. In (7c), on the other hand, ho ‘pigs’ is animate and a count
noun and therefore triggers acc agreement -ad ‘them’ on the verbs qoc ‘hit’
and jec ‘eat’.6 In this case the object is also specific and ‘eat’ is an active
accomplishment.
25.2.2.1 Verbs
There are six morphosyntactic forms of the verb: regular verb, impersonal
verb, reciprocal verb, light verb, serial verb, and dependent switch-reference
verb. Serial verbs are described in Section 25.3.1. They are not marked for
tense or for nom agreement. Dependent switch-reference verbs are
described in Section 25.3.2. They are not marked for tense either. Amele
also has an interrogative verb adec ‘to when’, as illustrated in (12).
25.2.2.2 Nouns
Amele has three morphological classes of nouns: regular nouns with
uninflected stems, inalienably possessed nouns with possessor agreement
inflection, and deverbal nouns, derived from the infinitive form of a verb.
Inalienably possessed nouns are described in Roberts (1987: 171–175, 2015b).
They comprise kin terms,8 body-part terms and personal attribute terms.
Deverbal nouns are formed from the infinitive form of the verb. The infini-
tive suffix, -ec/-oc, on the verb functions as a nominalizing suffix on the noun
(e.g. cob-oc [walk-inf] ‘to walk’ or [walk-nmlz] ‘a walk’).
25.2.2.3 Postpositions
Postpositions are an important minor word class in Amele (Roberts 1987:
160–161). They are clitic words which must attach to a preceding host
element and they may not be stranded (Roberts 1991c, 1992, 1996). They
can be predicative or non-predicative, and we return to this in Section
25.2.4.2.
SENTENCE
CLAUSE
(PERIPHERY)
CORE
PRED
V DUn DN PSA
Figure 25.2 The layered structure of the clause
SENTENCE SENTENCE
CLAUSE CLAUSE
Templates for optional syntactic structures are given in Figure 25.3. The
pre-detached position (PrDP) is for topical established information and the
post-detached position (PoDP) is for additional information, such as clarifi-
cation. The pre-core slot (PrCS) is for focal new information or ‘heavy’
constructions, such as a nominal modified by a relative clause. The post-
core slot (PoCS) is for postposed clausal elements.
PrDP example
Example (14) illustrates a clause-external topic caja eu ‘that woman’ in the
pre-detached position. There is a resumptive pronoun uqa ‘she’ in the clause
for the argument in the PrDP.
PoDP example
In (15) the RP dana eu ‘that man’ in the PoDP adds clarifying information to
the identity of the PSA eu ‘that’ in the clause. This pronoun is resumptive for
the argument in the PoDP.
PrCS examples
In the unmarked form, temporal adjuncts (RP or PP) occur after the PSA RP,
as in (16a). Alternatively, the temporal adjunct can be placed in the PrCS as
focal new information, as in (16b).
Arguments and adjuncts with a relative clause are typically placed in the
PrCS, as shown in (17).
PoCS example
In (18) there is an instrumental argument-adjunct PP dubin¼na ‘with
stalk’ and a goal argument-adjunct PP camac ta¼na ‘into the sago scrapings’
in the final clause. The second argument-adjunct PP is located in the PoCS.
25.2.3.2 Operators
Illocutionary force
Illocutionary force (IF) is the outermost operator and the form of expression
of different types of IF in Amele is illustrated in (19)–(23). Statement (19) is
expressed by a declarative sentence. Command (20) is expressed by the
imperative form of the verb. The imperative is identical in form to the
today’s past tense form. A yes/no question (21) is expressed with the
sentence-final question particle ¼fo. An information question (22) is
expressed by an interrogative word, such as in ‘who’, eeta ‘what’, cel ‘which’.
Exhortation (23) is expressed by the hortative form of the verb.
Evidentials
Amele does not have evidentials.
Status
Amele expresses realis/irrealis status on the switch-reference verb (see
Roberts 1990, 1994 for a wider study of this phenomenon in Papuan lan-
guages). The simultaneous ds nom agreement has different forms
depending on the realis/irrealis status of the clausal operator category
marked on the final verb in the linked clauses, traditionally called a clause
chain in Papuan linguistics. The clausal operator categories with realis
status are: present tense, today’s past tense, yesterday’s past tense, remote
past tense, habitual past tense/aspect, negative past tense. The clausal
operator categories with irrealis status are: future tense, prospective (about
to) tense, negative future tense, and imperative, prohibitive, hortative,
optative, counterfactual, apprehensive IF categories. An example of realis
status is given in (24a) and an example of irrealis status is given in (24b).
Tense
The tense categories in Amele are metrical. Metrical tenses mark degrees of
temporal remoteness from the deictic centre. Amele has a present tense (3a)
and three degrees of past tense: today’s past tense (3c), yesterday’s past tense
(22), and remote past tense (3b), as well as a negative past tense and a past
habitual tense/aspect (11b′).10 The language has a regular future tense (25), a
negative future tense (38), and a prospective (about to/intentional) (26)
tense.11 The event referred to in the prospective tense is nearer to the deictic
centre than the event referred to in the regular future tense. It is therefore
metrical.
Modality
Categories of deontic modality, such as ability, permission and obligation
are expressed lexically in Amele (Roberts 1987, 2001).
Event quantification
Event quantification (Roberts 2015a) is marked on the verb by distributive
inflection that is homonymous12 with the acc agreement, -ad ‘plural’ and -al
‘dual’. Some examples are given in (27). In (27a) the motion verb belec ‘to go’
is intransitive and the -ad ‘plural’ marker indicates a multiple event of each
woman going her own way. In (27b) and (c) calec ‘to arrive’ is intransitive. The
-ad ‘plural’ marker in (27b) indicates multiple events of men arriving, while
in (27c) the -al ‘dual’ marker indicates two events of men arriving.
The EVQ marker -ad can also indicate an exclusive action, as in (28).
Aspect
There are three types of aspect that can be expressed by reduplicating some
part of the verb word: durative aspect, regular and irregular iterative aspect.
Durative aspect is marked on the simultaneous ss/ds verb and conveys
the idea that there is an extended temporal overlap of events. Without this
marking the temporal overlap is punctiliar. Durative aspect is expressed by
different kinds of CV~ or V~ reduplication of the verb stem (Roberts 1991a).
See (24a, b), (28), for examples of CV~ reduplication and (12) for an example
of V~ reduplication. With some verbs, the reduplicative (C)V~ marking of
durative aspect applies to the nom agreement suffixation rather than to the
verb stem, as illustrated by (29). If there is accusative marking on the verb
then the reduplicative marking for durative aspect applies there, as in (30).
does not have an acc marker, otherwise the acc marker is reduplicated
either in place of, or in addition to the reduplication of the verb stem
(Roberts 1987: 252–256). For some minimal stem verbs, such as l-ec ‘to go’,
the reduplicated stem is of the serial verb form (e.g. li~li-ec ‘to go repeatedly’;
see Section 25.3.2). An example of regular iterative aspect is given in (31a).
ebum oso ‘one hand (10)’. Ordinal numbers can be formed by adding the
nominalizing suffix -doc to the numeral (e.g. osahic-doc ‘first’, lecis-doc ‘second’,
cijed-doc ‘third’). There are also a number of relationship terms that can
function as ordinal numbers, for instance matu ‘firstborn/first’, milum
‘secondborn/second’, subig ‘lastborn/last’.
The quantifier can be followed by a deictic element, such as eu ‘that’ or
ceheleg ‘up there’, or the indefinite article oso ‘a/one’. The article can co-occur
with a deictic element, as in dana eu oso [man that art] ‘one of those men’.
The universal quantifier cunug ‘all’ occurs at the end of the RP, as in jo nag
cijed eu cunug [house small three that all] ‘all those three small houses’.
With respect to (32a), the head noun can be questioned with an interroga-
tive, such as eeta ‘what’ or in ‘who’, for example Hina ija-in in? [2sg name-2sg.
psr who.sg] ‘What (lit. who) is your name?’ The numeric quantifier can be
questioned with an interrogative, such as ganic ‘how much/many’, for
example jo ganic? [house how many] ‘how many houses?’, saab ganic? [food
how much] ‘how much food?’. However, when the deictic is questioned, the
interrogative form, such as cel ‘which’, is placed in the RPIP, which is the pre-
core slot in the RP, as in (32b). Thus the interrogative form is cel dana [which
man] ‘which man?’.
With respect to (32c), the preceding PP/RP expresses the possessor of the
noun or a specification of the noun. (33) illustrates the difference in how
alienable and inalienable possession is expressed.14 Alienable possession in
(33a) is expressed with a possessive PP preceding the possessed noun. The
possessive PP functions as an argument of the possessed noun, which is the
head of the nucleus of the possessive noun phrase. The possessive PP is
therefore non-predicative. Inalienable possession in (33b) is expressed by
suffixal agreement in person (first, second or third) and number (singular,
dual or plural) with the possessor noun phrase. The possessor need not be
expressed by an overt RP or pronoun. Thus the possessor agreement func-
tions as the argument (ARG) of the inalienably possessed head noun.15
Semantically, inalienably possessed nouns are kinship terms, body-part
terms or personal attributes. In (33b) cuduni means ‘my personal place’ or
‘the place that belongs to me in some way’, whereas in (33a) ijana cudun
simply means ‘my place’ without the connotations of personal ownership.
When the modifier follows the head noun it has an attributive function.
In (34a) jo us nijec describes a house where people are sleeping. When the
RP MP
CORER MP
NUCR MP
N N NUM DEM
NUCR
CORER QNT
RP DEF
RP DEIC
Figure 25.4 The basic structure of the RP in Amele
RP
CORER
PP NUCR
RP P
RPIP CORER
RP NUCR ARG
PRO N N
25.2.4.2 PP Structure
Amele only has postpositional phrases (PPs). They can be predicative or
non-predicative. Predicative PPs occur with temporal or locative adjuncts
(e.g. Mande¼na [Monday¼on] ‘on Monday’, be-on′ (Mande, x) or jo¼na
[house¼in] ‘in the house’ be-in′ (jo, x)), or with goal, path, source, or
instrument argument-adjuncts, as described in Section 25.2.7. The
possessor-marking postposition ¼na ‘of’, described in Section 25.2.4.1, is
non-predicative.
directly to the verb stem and the perceived argument cannot be marked on
the verb.
The verb ‘give’ is exceptional in that the undergoer argument func-
tions as the stem of the verb. The first argument of have′ (x, y) (recipient)
in ‘give’ LS is thus the only choice for undergoer assignment and there is
no alternative argument-marking construction where the second argu-
ment of have′ (x, y) (theme) is the undergoer. An example of this was
given in (10).
When the goal argument in an argument-adjunct PP is a person, an
applied object construction may be used as an alternative expression, as
illustrated in (4). This is also the case for the source adjunct PP argument,
and the benefactive adjunct PP argument, illustrated in (37). For the mal-
efactive argument shown in (38) there is only the applicative form and there
is no corresponding argument-adjunct PP form.
SENTENCE
PrDP CLAUSE
CORE
PRED
‘That woman, she is no good.’
N
IU IU IU
SPEECH ACT
Figure 25.6 The PrDP and the potential focus domain
SENTENCE
CLAUSE PoDP
RP CORE
PP NUC ARG RP
PRED
PRO V
IU IU IU IU
SPEECH ACT
Figure 25.7 The PoDP and the potential focus domain
man’. The actual focus domain is the predication uqa jejegca men ‘got his
voice back’.
Amele has a range of particles that can attach to the end of the sentence
which qualify the proposition, such as da ‘but, however, nevertheless’, do ‘be
encouraged, let’s do it’, fo ‘yes/no question’ fa ‘dubitive question, maybe’, ijom
‘certainly’, le ‘permission granted’, and mo ‘supplication, pleading’ (Roberts
1990). Some of the sentence particles, such as the question particles, fo and fa,
can occur either at the end of the sentence or they can be focused on a
particular constituent in the sentence. In (40a) the yes/no question particle,
fo, occurs at the end of the sentence. It has scope over the whole proposition
and therefore expresses sentence focus. In (40b–c) the question particle is
focused on a particular sentence constituent and expresses marked narrow
focus on that constituent. (40d) illustrates the use of the question word in
‘who’. It is ungrammatical to have the question particle fo at the end of the
sentence in this case since the question word expresses narrow focus and the
particle at the end of the sentence would express sentence focus. However, as
illustrated in (40e), it is possible to focus on the question word with the
question particle. This expresses marked narrow focus.
25.2.10 Linking
In this section the semantics-to-syntax linking for some two-place and three-
place predicates is illustrated. Figure 25.8 shows the semantics-to-syntax
linking in a transitive sentence with acc (DUn) marking. Here 3pl [dana]
is the first argument in the active predicate LS do′ (3pl, [eat′ (3pl [dana], 3pl
[ho])]), and is assigned the actor macrorole. 3pl [ho] ‘pigs’ is the second
argument in this LS and also the only argument in the stative predicate LS
ingr consumed′ (3pl [ho]). This argument is assigned the undergoer macro-
role. The actor argument, 3pl [dana], is selected as PSA and coded as nom
agreement on the verb. The undergoer argument, 3pl [ho], is coded as
acc agreement on the verb in order to indicate the plurality of the
animate referent.
Figure 25.9 illustrates the semantics-to-syntax linking in an IVC. The 1sg
argument in feel′ (1sg, [afraid′]) is assigned the undergoer macrorole
according to the Actor-Undergoer Hierarchy (AUH). The IVC is coded as
3sg.nom agreement by default. There is therefore no linking between this
argument marking in the syntax and the logical structure.
Figure 25.10 illustrates the semantics-to-syntax linking for a three-place
predicate, ihacdoc ‘to show him’. The leftmost argument, 3sg ‘he’, is
assigned the actor macrorole and the rightmost argument, 3pl [ho] ‘pigs’,
SENTENCE
CLAUSE
RP RP CORE
PRED
‘Those men … ate the pigs.
N PRO N V
ACTOR UNDERGOER
[ ′ (3PL [dana], [ ′ (3PL [dana], 3PL [ho])]) … ] & [INGR ′ (3PL [ho])]
Figure 25.8 Semantics-to-syntax linking in a transitive sentence with ACC marking
SENTENCE
CLAUSE
RP CORE
NUC ARG
PRED
‘I am afraid.’
PRO V
UNDERGOER
is assigned the undergoer macrorole. The actor is selected as PSA and coded
as nom agreement on the verb. The undergoer is coded as a DUn by acc
agreement attaching directly to the verb stem. The non-macrorole core
argument, 1sg ‘me’, is coded as a DN by the applicative marker and acc
agreement marking on the verb.
SENTENCE
CLAUSE
RP CORE
PRED
‘He showed those pigs to me.’
N V
(9b) Ho eu ihac-ad-i-t-ei-a.
pig that show-3PL.ACC-APPL-1SG.ACC-3SG.NOM-TODP
The language-specific topics discussed in this chapter are serial verb con-
structions in Section 25.3.1 and switch-reference in Section 25.3.2.
g. Obligatory non-coreference: a true SVC will not contain two overt NPs
which refer to the same argument.
h. A prototypical SVC contains only one grammatical subject.
SVCs in Amele do not have all of the prototypical properties listed in (42).
In an SVC the non-final verb in the series is marked with -i or -u. Verbs with
an -ec infinitive form take -i, while verbs with an -oc infinitive form take -u.
The -i/-u marking indicates the verb is dependent (dv).18 This contravenes
principle (42b) as the dv marker indicates a nexus relationship. The PSAs of
the verbs in an SVC typically have the same referent.19 There are two basic
types of SVC. In one type, illustrated in (43a), the verbs are in a cosubordi-
nate relationship. In the other type, illustrated in (44a), the verbs are in a
superordinate–subordinate relationship. By comparison, the verb manimei in
(43b) is fully inflected with switch-reference morphology and is therefore
not a serial verb. Similarly, ehimeig in (44b) is not a serial verb.
In (43a) the verbs mani ‘roast’ and jeia ‘she ate’ describe a series of closely
related events. Compare (43b) in which manimei is the fully inflected ss.seq
form. Here the roasting and eating are interpreted as separate consecutive
events and they are separate clauses in the syntactic representation. The
arguments caja ‘woman’ and ceta ‘yam’ in (43a) are shared in the LS for
the two predicates. In the syntax for (43a) illustrated in Figure 25.11 the
argument caja is represented by an RP in the clause and by 3sg.nom
SENTENCE
CLAUSE
RP CORE
CORE CORE
PRED PRED
agreement on the verb jeia ‘she ate’. The argument ceta is represented as a
core RP argument of mani ‘roast’. It is not coded on the verb by acc
agreement as it is inanimate and a mass noun. The SVC is a cosubordinate
CORE [[CORE] [CORE]] juncture because it is not possible to have temporal or
locative clausal adjuncts applying separately to either core. By comparison,
in (43b) it would be possible to qualify jeia ‘she ate’ with the temporal
adjunct hibna ‘later’ because it is a separate clause.
(44a) describes a single event of ‘take down’. The verb n- ‘come down’ is a
directional modifier of the verb ehi ‘take’ and is coded in the LS as such.
There is no shared argument between the verbs as n- is represented in the LS
as a directional operator. (44a) therefore contravenes principle (42f). Com-
pare (44b), where ehimeig is fully inflected for ss.seq. Here the event ehimeig
‘they took’ is linked to the event neiga ‘they came down’ by {&} in the LS.
They are separate consecutive events in the LS and are separate clauses in
the syntactic structure. In the syntactic structure for (44a) in Figure 25.12 the
verb n- is in a subordinate relationship to ehi. This is ad-nuclear
subordination. The 3pl.nom argument agreement -eig attaches to n- as the
final verb in the series.
With respect to the characteristic SVC properties detailed in (42), a par-
ticular SVC may not necessarily belong to a single intonation contour
(cf. 42c). Compare the sentence in Figure 25.13. The li ‘go’ verb is a direc-
tional modifier of ehi ‘take’. The first two cores form one intonational unit
and the last core is a separate intonational unit. However, the dv coding
marks this as describing a series of linked events which should be inter-
preted as a unified complex event.
The subordinate SVC has a range of modifying functions in addition to
expressing directionality. Examples are given in (45)–(49). In each case the
modifying verb follows the verb that is modified. (45)–(46) express aspect
periphrastically. (45) illustrates how continuative aspect is expressed by a
SENTENCE
CLAUSE
RP CORE
PRED
‘They brought the firewood down.’
PRO N V V
SENTENCE
CLAUSE
N PRO N V V V N V
posture verb: bilec ‘to sit’, nijec ‘to lie’, tawec ‘to stand’, and (46) illustrates
how completive aspect is expressed with the verb hedoc ‘to finish it’. (47)–(49)
illustrate some other common forms of this type of subordinate SVC.
(46) Completive aspect expressed with the verb hedoc ‘to finish it’:
Age jo ceh-i he-d-oig-a.
3pl house plant-dv finish-3sg.acc-3pl.nom-todp
‘They finished building the house.
(47) Investigative modification expressed with the verb fec ‘to see’:
Uqa wehuc j-i f-ei-a.
3sg soup eat-dv see-3sg.nom-todp
‘She tasted the soup.’
(48) Superlative modification expressed with the verb cuhadoc ‘to excel’:
Caja uqa ola-Ø cus-i cuha-d-on.
woman 3sg face-3sg.psr scrub-dv excel-3sg.acc-3sg.nom.remp
‘The woman cleaned her face well.’
25.3.2 Switch-Reference
Switch-reference (SR) in Amele is judged to be a local syntactic device for
monitoring the referentiality of PSA arguments between adjacent clauses as
to whether they have identical or non-identical reference (Roberts (2017).
Dependent SR verbs occur most commonly in a clause chain.22 A clause chain
comprises a string of linked clauses with only the final clause in the chain
marked for clausal operator categories, such as tense or illocutionary force.
The non-final clauses are dependent on the final clause for clausal operator
designation. The operator-dependent clauses are also typically marked with
switch-reference morphology. Amele has two basic types of ss/ds morphology.
One codes sequential events and the other codes simultaneous overlap of
events. The sequential verb is marked with either -im ‘ss.seq’ or -ec/-oc ‘ds.
seq’ followed by nom agreement morphology. The simultaneous verb is
marked with nom agreement morphology which indicates either ss.sim or
ds.sim. The ds.sim morphology is further divided into that which indicates
realis status and that which indicates irrealis status of the tense or IF
category marked on the final clause in the clause chain.23 The ds.sim verb
agrees in status value with the status value of the final clause. ss/ds clauses
are most commonly in a cosubordinate relationship with other clauses (see
Chapter 13). However, ss/ds clauses can also have a subordinate function.
Note that the final clause in a clause chain need not be finite. In (51) the
purpose clause ho bubusaleb qoc ‘to kill the pig as it runs out’ is a clause chain
with qoc ‘to kill’ as the final clause. The verb bubusaleb in the dependent
clause is marked for ds simultaneous irrealis nom agreement morphology
as the understood PSA of qoc is ‘those men’, and qoc, being infinitive, has
irrealis status.
SENTENCE
CLAUSE
PP NUC ARGDS
PRO V V V V V
Ija Malolo uqana ka jic anag ononu sumudi bi~bil-igin n-ec-eb tob-oc-omin bel-owan=Ø.
SENTENCE
Figure 25.14 Cosubordinate clause chain structure, example (50)
overlap between the pivot referents of the controlling clause and the pivot
referents of the dependent clause. Where the pivot referent(s) of the control-
ling clause is (are) properly included in the set of pivot referents of the
dependent clause then ss is marked. However, when the pivot referent(s) of
the dependent clause is (are) properly included in the set of pivot referent(s)
of the controlling clause then ds must be marked. For example, in (50) 1sg.
nom in [3] is properly included in 1du.nom in [4] and in this case ds must
be marked. If it was the other way around (‘we (du) climbed in and off
I went’) and 1sg.nom was the controlling referent and 1du.nom was the
dependent referent then ss would be marked.25
Stirling (1993) suggests that SR in Amele tracks events rather than the main
participant(s) in the event. As already mentioned, Amele has asymmetrical
SENTENCE
CLAUSE
RP CORE
RP CORE
NUC ARG
PRED PRED
N N V V
NUC NUC
CORE
CLAUSE
CORE
SENTENCE
Figure 25.15 Core subordinate DUn clause, example (56a)
SENTENCE
CLAUSE
CLAUSEDS CLAUSEDS
Ceta bahim mi hedumei ceta wal meceb ceta eu hunimei gulden hocob ceta bahimna tacen=Ø.
SENTENCE
Figure 25.16 Layered cosubordination in (57)
The PSA in the sub-cosubordinate clause is different to the PSA in the matrix
clause in each case, so ds is marked.
Note that this produces an anomalous SR marking between clause [1] and
the immediately following clause [2], and between clause [3] and the imme-
diately following clause [4]. The verb in clause [1] is marked for ss when the
PSA of clause [2] is different to the PSA in clause [1]. The verb in clause [3] is
also marked for ss when the PSA of clause [4] is different to the PSA in clause
[3]. This shows that the switch-reference marking system takes account of
the layers of cosubordination. For clauses signalling the mainline events,
such as clauses [1, 3, 5] in (57) and the four clauses in (50) the SR-marking
system selects the next cosubordinate clause as the controlling clause for SR
marking. For an SR clause indicating background information, such as [2, 4]
in (57), it is interpreted being relevant to the following ‘mainline event’
expressing clause, and the SR-marking system selects that clause as the
Now compare (60b). Here, cucuiteceb ‘I was afraid’ is an IVC.28 In this case,
the verb has a 1sg.acc ‘experiencer’ argument and a 3sg.nom ‘empty’
argument. When cucuiteceb functions as the controlling clause to ija cocobig
this verb is marked ss. Thus, the 1sg.acc argument in the IVC functions as
pivot when it is the controlling clause. However, when cucuiteceb functions as
the dependent clause to busali nuiga it is marked ds. This shows that the
pivot arguments are not being compared between these clauses. Instead, the
controller 3sg.nom argument in the IVC is compared with the pivot in the
non-IVC controlling clause. Thus, the coreferentiality rule in (53) needs to be
modified to (61). (61) applies to both regular verbs and IVCs. With a regular
verb the pivot and controller are one and the same argument so monitoring
either function does not select a different argument. However, in an IVC, the
pivot and controller are different arguments, so it makes a difference which
argument is monitored when the IVC functions as controlling clause or
dependent clause.
References
Roberts, John R. 1994. The category ‘irrealis’ in Papuan medial verbs. Notes on
Linguistics 67: 5–39.
Roberts, John R. 1996. A Government and Binding analysis of the verb in
Amele. Language and Linguistics in Melanesia 27(1): 1–66.
Roberts, John R. 1997a. Switch-reference in Papua New Guinea:
A preliminary survey. In A. Pawley (ed.), Papers in Papuan Linguistics,
No. 3, 101–241. Canberra: The Australian National University.
Roberts, John R. 1997b. The syntax of discourse structure. Notes on Translation
11(2): 15–34.
Roberts, John R. 1997c. GIVE in Amele. In John Newman (ed.), The Linguistics of
Giving, 1–33. Amsterdam: John Benjamins. www.sil.org/resources/archives/
22980.
Roberts, John R. 2001. Impersonal constructions in Amele. In Alexandra Y.
Aikenvald, R. M. W. Dixon and Masayuki Onishi (eds.), Non-Canonical
Marking of Subjects and Objects (Typological Studies in Language 46),
201–250. Amsterdam: John Benjamins.
Roberts, John R. 2012. Serial verbs in English: An RRG analysis of catenative
verb constructions. Functions of Language 19(2): 201–234.
Roberts, John R. 2015a. Distributives in Amele. A Role and Reference Gram-
mar analysis. SIL Electronic Workpapers 2015–001. www.sil.org/resources/
publications/entry/60478.
Roberts, John R. 2015b. Inalienable possession in Amele: A Role and Refer-
ence Grammar account. SIL Electronic Workpapers 2015–002. www.sil.org/
resources/publications/entry/60480.
Roberts, John R. 2017. The typology of switch-reference. In Alexandra Y.
Aikhenvald and R. M. W. Dixon (eds.), Handbook of Linguistics Typology,
538–573. Cambridge: Cambridge University Press.
Stirling, Lesley. 1993. Switch-Reference and Discourse Representation. Cambridge:
Cambridge University Press.
Talmy, Leonard. 2007. Lexical typologies. In Timothy Shopen (ed.), Language
Typology and Syntactic Description, 66–168. Cambridge: Cambridge University
Press.
Toratani, Kiyoko. 2002. The Morphosyntax and the Logical Structures of Compound
Verbs in Japanese. Unpublished PhD dissertation, University at Buffalo (SUNY).
Z’graggen, John A. 1975. The Languages of the Madang District, Papua New Guinea
(Pacific Linguistics B-41). Canberra: The Australian National University.
Notes
1 See Roberts (1987) for a full linguistic description of the Amele language
from a typological perspective. The dialect of Amele represented in this
study is the Haija dialect (Roberts 1991b).
Abbreviations
The abbreviations and symbols used in the glosses in general follow the
Leipzig glossing rules (www.eva.mpg.de/lingua/resources/glossing-rules.
php), with the following additions:
26.1 Introduction
This chapter presents a sketch grammar of Amis with a focus on the discus-
sion of the phenomena related to its case marking and voice. Amis is
currently spoken by 210,839 speakers in Taiwan.1 It is classified as one of
the Eastern Formosan branch languages (Blust 1999) and is reported to
consist of five dialects (Tsuchida 1988). The following discussion is based
on the data collected from the Central dialect.
As shown in (1), verbal predicates are usually affixed with a voice marker
(e.g. mi-), followed by various numbers of arguments that are marked by a
case marker, be it a nominative case (for the privileged syntactic argument
(PSA)), a genitive case (for a non-PSA actor or a possessor), or a dative case (for
an oblique core argument or an adjunct). A locative argument or adjunct is
marked by the preposition i.
Nominal predicates and prepositional predicates are exemplified respect-
ively in (2a) and (2b).
a. O singsi cingra.
cn teacher 3s.nom
‘He is a teacher.’
a′. be′ (cingra, [teacher′])
b. I loma’ Ø-ci Mayaw.
prep house nom-ppn Mayaw
‘Mayaw is at home.’
b′. be-at′ (loma’, Mayaw)
Both of the constructions contain a displaced element that appears before the
main predicate, but there is no pause before the element in the displacement
construction and the rest of the clause while there is one in the topic construc-
tion, which means that the displaced elements in (3a) and (3b) are placed
under the pre-core slot (PrCS) and pre-detached position (PrDP) respectively.
The discussion of lexical categories includes the roots and derived words in
Amis. As pointed out by Wang (1976), all the root forms in Amis are syntactic-
ally nominal.5 In other words, to serve as a verb, root forms, even for roots
with inherent verbal meaning, have to be either affixed with a verbal prefix
(usually a voice marker6) or placed in the predicate position. The former
strategy is more commonly found and examples can be seen in (1); the latter
is only found with some state predicates such as miming ‘small’ and tosa ‘two’.
Derived verbs with no verbal affixes are termed unaffixed verbs in Wu (2006a).
As root forms are syntactically nominal, they can readily appear after a case
marker or a preposition (e.g. (4a)) to manifest arguments and adjuncts. How-
ever, the derived verbs (e.g. root form plus a voice marker) have to be nominal-
ized7 before they can show up in a nominal position (e.g. (4b–c))8. See Wu
(2006a: 70) for some common nominalizing or deverbal strategies in Amis.
a. Ma-’orad anini.
neut-rain now
‘It is raining today.’
a′. rain′ (Ø)
b. Ma-’orad k-o kakarayan.
neut-rain nom-cn sky
‘The sky is raining.’
b′. rain′ (kakarayan)
Regarding the assignment of the macrorole for verbs with one core argu-
ment (i.e. syntactically intransitive or S-intransitive verbs), the presence or
absence of do′ in their logical structure (LS) makes a crucial difference; the
presence of do′ makes the only core argument an actor while the absence of
do′ makes it an undergoer. In other words, the single argument for intransi-
tive activity verbs such as t<om>ireng ‘stand’ and intransitive state verbs
such as ma-toni’ ‘soft’ will not be assigned the same macrorole:
a. T<om>ireng cingra.
<neut>body 3s.nom
‘He is standing.’
b. Ma-toni’ k-o-ni a titi.
neut-soft nom-cn-this lnk meat
‘This meat is soft.’
Although the only argument of the two verbs in (6) ia marked by the same
case (i.e. the nominative case), the macrorole of t<om>ireng would be actor
while that of ma-toni’ would be undergoer. This is where the Role and
Reference Grammar (RRG) analysis differs from other studies concerning
this issue in Amis (e.g. Liu 1999, Liu 2003, to name just a few). In such
research, the single arguments of S-intransitive verbs seem to be treated as
the same type and the verbs are labelled as actor-focus or actor-voice verbs.
This analysis will be further discussed in the last section of this chapter.
When there are two core arguments in the LS of an Amis predicate, the
macrorole assignment follows the Actor-Undergoer Hierarchy (AUH) and the
default macrorole assignment principles. However, it is also possible for
such verbs to have only one macrorole. As illustrated in (1b) and (1c), two-
place verbs can appear with two case-marking patterns: the nominative-
dative case frame in the actor voice (AV) sentences and the genitive-
nominative pattern in the undergoer voice (UV) sentences. While the two-
place UV verb follows the AUH and has both an actor and an undergoer, the
second argument in their AV counterparts is a non-macrorole core argu-
ment, as argued in Wu (2006a). In other words, the two-place AV verbs are M-
intransitive. The non-macrorole status of the second argument results either
from the linking phase from the argument position in the LS to macrorole
or from the voice operation that deprives a core argument of its macrorole
status during linking from macrorole to syntactic function. The first case is
found when the second argument of an activity verb is non-referential, as
exemplified in (7a). The second one is illustrated in (7b).
As shown in (8), both the recipient argument and the theme argument can
be the PSA,9 in particular, the syntactic pivot, of the applicative three-place
verb pa-nanom-an ‘give water’. This is a property for a non-macrorole (NMR)
argument. In other words, neither of the two non-actor arguments in AV pa-
sentences such as (8a) are macroroles.
While the UV three-place predicates are M-transitive, just like the UV two-
place UV predicates, the mismatch between S-transitivity and M-transitivity is
still found as there can be at most two macroroles in RRG, while there are three
core arguments in a three-place predicate. This means there is competition for
macrorole status between the two groups of potential undergoer participants,
namely, theme/patient and recipient/beneficiary/source/goal. Hence, two prin-
ciples of undergoer selection (i.e. Principle A: choosing the lowest-ranking
argument in LS and Principle B: choosing the second-highest-ranking argu-
ment in LS) have been proposed in RRG under the AUH. As reported in Guerrero
Valenzuela and Van Valin (2004), languages tend to exhibit a mixed type that
needs both principles to account for their undergoer selection. Amis is also
such a language. Wu (2006a, b) has shown that Amis three-place predicates vary
in their selection of the undergoer, though Principle B seems to be more
commonly employed. The evidence lies in the possibility of having either the
theme/patient or the recipient/beneficiary/source/goal argument as the PSA in
the UV constructions, and the latter group seems to be favoured by more three-
place predicates.10 Some examples are given below. As one can see, verbs like
pa-caliw ‘lend’ can only have the theme argument as the PSA in their UV
construction, while verbs like pa-nanom ‘give water’ only allow the recipient
argument as the PSA in their UV counterpart.
After briefly reviewing how the macroroles are selected for verbs with
various numbers of core arguments, now let us see how these arguments
and non-arguments are marked in Amis.
26.4.2 Argument-Marking
Arguments and non-arguments are marked in Amis through a tri-case
system (nominative, genitive, dative) and the preposition i. The nominative
case marks the PSA, or the so-called grammatical subject. The genitive case
can mark a possessor or an actor in a non-AV sentence (e.g. (1c)), which
makes it the equivalent marker of an ergative case. The dative case serves a
wide range of functions. It can label a NMR core argument, as seen in the
above examples (e.g. (1b)). It can also appear before an oblique core argu-
ment or an adjunct. Examples follow.12
The dative case marks an oblique core argument in (10a) and various kinds of
adjuncts in (10b–d). The oblique status of the theme argument in (10a) is
The data in (11) illustrates the possibilities for enhancing the status of the
second argument of pred′ (the one marked by the dative case) in mi-nanom
and ma-olah to become a PSA (i.e. undergoer of a UV verb). Both the plain UV
constructions (e.g. (11a) and (11c)) and the applicative UV constructions (e.g.
(11b) and (11d)) are applicable here. Note that the number of core arguments
in the two predicates has remained the same in the plain UV constructions
and the applicative UV constructions.13
Now consider a different case exemplified in (12).
b′. [do′ (mato’asay, [use′ (mato’asay, sapaiyo no ’edo)])] CAUSE [[do′ (sapaiyo no
’edo, Ø) ]CAUSE [BECOME dead′ (’oner)]]
c. Ma-ota’ kako t-o sanek n-o tosiya.
neut-vomit 1s.nom dat-cn smell gen-cn car.
‘I feel like vomiting from the smell of cars.’
c′. do′ (kako, [vomit′ (kako, (y))])
d. Sa-ka-ota’ ako k-o sanek n-o tosiya.
ia-nfin-vomit 1s.gen nom-cn smell gen-cn car
‘The smell of the car is the reason why I vomit.’ (Instrumental applicative, UV)
d′. because.of′ (sanek no tosiya, [do′ (kako, [vomit′ (kako, (y))])
The reason/indirect cause adjunct RPs in (12a) and (12c) now become the actors
in (13a) and (13b), respectively. As indicated in the logical structures of mi-patay
in (13a) and mi-ota’ in (13b), the predicates have become causativized, and there
is an effector added to the core of the predicates.14 In other words, the number
of core arguments has also been changed. The addition of a core argument is
not found in the examples in (11); when the to RPs of mi-nanom and ma-olah
become PSAs in the plain or applicative UV constructions, there is no argu-
ment addition involved. Hence, the to RP of ma-patay in (12a) and ma-ota’ in
(12c) should be analysed differently from the to RPs in mi-nanom in and ma-olah;
the former are adjuncts while the latter are NMR core arguments.
The last difference between an NMR direct core argument and an
adjunct (or even an oblique core argument) is that the status of the
former can always be adjusted through the plain voice operation; how-
ever, for the latter, it is not always possible. In other words, some to RPs
can only be promoted by means of the applicative constructions. For
example, the plain voice construction is quite unlikely to be employed to
promote the adjunct manifesting temporal expression in (10c), although the
applicative form ka-tayal-an ‘place or time for working’ can be used. For some
adjuncts that are more likely to be construed as effectors (e.g. an indirect cause
like sanek no tosiya ‘smell of the car’ in (12c), they may be promoted to become
an actor in AV and UV constructions, as we have seen in (13).
Based on the discussion so far, the case assignment rules in Amis can be
summarized as (14):15
The examples in (15) illustrate how the rules in (14) are applied in Amis.
As indicated in (17), the first argument of the embedded be-loc′ (e.g. langa
‘basket’)) cannot be an undergoer in the UV construction. This follows
from the claim in RRG that the first argument of be-loc′ or pred-loc′
Assign the preposition i to the first argument of . . .pred′ (x, y). . . if it is a non-
macrorole argument:
(i) obligatory if pred′ (x, y) ¼ pred-loc′ (x, y), x ¼ common noun
(ii) optional if pred′ (x, y) ¼ pred-loc′ (x, y), x ¼ personal proper noun
(iii) optional if pred′ (x, y), pred′ ¼ cognition, possession, and perception
This section discusses grammatical relations and the voice system in Amis.
According to Wu (2006a), Amis only employs a PSA in the formation of
relative clauses, a nominal type of displacement constructions and wh-ques-
tions. Constructions that might need a grammatical relation in other lan-
guages, such as control constructions and reflexivization, can be taken care
of by semantic roles in Amis. Hence there are no grammatical relations
found in such constructions.
As for the voice system, Amis distinguishes two voices: actor and under-
goer, and the latter can be further divided into the plain (or non-applicative)
UV and the applicative UV. The functions of each voice will be elaborated in
the following sub-sections.
Table 26.1 Voice markers and the applicative markers in Amis (adapted from Wu
2006b: 289)
Let us begin with the discussion of actor voice. AV verbs always have a
nominative-dative case pattern, and they have only one macrorole (i.e.
actor). The actor is assigned the nominative case and the NMR argument is
marked by the dative case. The voice-marking function of the AV affixes is
demonstrated in their co-occurrence with the volitative mood marker -aw,
which shows the UV pattern when it attaches to a root form.
a. Nanom-aw¼ho¼ako.
water-vol¼asp¼1s.gen
‘I will go drink water first.’ (Volitative mood, UV)
a′. Mi-nanom-aw¼ho kako.
av-water-vol¼asp 1s.nom
‘I will go drink water first.’ (The water is farther away than the one
mentioned in (23a).)
b. Kaen-aw¼ako k-o dateng.
eat-vol¼1s.gen nom-cn vegetable
‘I will try that vegetable.’ (Volitative mood, UV)
b′. K<om>aen-aw k-o wawa t-o sapaiyo.
<av>eat-vol nom-cn child dat-cn medicine
‘(I am) afraid that the child will take the medicine.’
c. Olah-aw¼ako kiso?
like-vol¼1s.gen 2s.nom
‘May I go to love you?’ (Volitative mood, UV)
c′. Ma-olah-aw kako tisonan.
av-like-vol 1s.nom 2s.dat
‘I am afraid that I will like you.’
As shown in (23), the suffix -aw manifests an optative reading for the derived
verb. Notice that the case-marking pattern for V-aw follows the UV pattern,
as the actor is marked by the genitive case. However, when the V-aw forms
are affixed with mi-, <om>, and ma-, their case-marking patterns become the
AV pattern, as indicated in (23a′), (23b′) and (23c′). This contrast shows the
voice-marking function of the AV markers. But what kinds of functions do
the AV constructions perform?
Clearly, the AV construction has a PSA-modulation function, as it makes a
marked choice of PSA in terms of the PSA Selection Hierarchy. Given the fact
that Amis displays ergative features in at least the case-marking system and
some constructions that involve a PSA such as the relative clause and the
nominal type of wh-question, one would expect the lowest-ranking argument
to be the unmarked PSA choice. However, in the AV construction, it is the
highest-ranking direct core argument that is chosen to be the PSA.
What about the argument modulation function? For a two-place predi-
cate, the lowest-ranking direct core argument in the AV sentences should be
assigned an undergoer based on the macrorole assignment principles, as
such verbs can take at most two macroroles. However, this argument in the
AV construction does not surface as a macrorole syntactically, as revealed by
its case marking and the fact that its status can be promoted by the appli-
cative construction. Instead, this argument is realized as an NMR core
argument in the AV construction. In other words, the lowest-ranking argu-
ment of a two-place predicate has been stripped of its macrorole status by
the AV operation. However, it is still in the core, as indicated by its behav-
ioural property in serving as a semantic controller in the persuade-type
control construction, as discussed in Wu (2006a). Hence, AV constructions
also perform an argument-modulation function. This function is even more
salient for three-place predicates, as a possible undergoer can also be
marked by the preposition in the AV construction in addition to the dative
case. Consider the following examples:
As shown in (24), the recipient RP singsi can be marked either by the dative
case or the preposition in the AV construction. This RP is the second-highest-
ranking argument in the LS of pa-caliw ‘lend’, and it is also a possible under-
goer, as indicated in the UV sentence in (24c). The presumed undergoer RP is
realized as non-macrorole in the AV construction in (24a), but it is realized as
an adjunct in (24c), as the preposition i typically marks a locative RP in the
periphery. From the above discussion, we can thus conclude that the actor
voice not only modulates the PSA choice but also modulates the semantic
status of a core argument by either stripping a macrorole argument of its
macrorolehood or realizing a core argument as an oblique element.
What about the undergoer voice? As shown in Table 26.1, there are two sets
of UV markers: the marked set and the unmarked set. The markers that
signify a marked undergoer choice used to be treated as voice markers as
well (termed instrumental voice and locative voice) in Amis, but the examples
in (25) show that they serve functions other than that of a voice operation.
The sentences in (25) show that when the UV marker -en or ma- and the
instrumental applicative marker sa- co-occur in a sentence, only the
instrument RP surfaces as the undergoer instead of the patient RP, which
would be the default undergoer following the AUH. In other words, the
applicative markers indicate a marked undergoer selection. The voice
markers and the applicative markers show different operations in the two
phases in the RRG linking algorithm. That is to say, the applicative marker
affects the linking from argument positions to macroroles, while the voice
marker operates at the linking from macroroles to syntactic functions.
There are two functions of these applicative markers. First, they may
enhance the status of a non-argument such as instrument or location to
become a core argument. Second, they can also promote a non-macrorole
core argument (e.g. patient in an AV sentence) to become a macrorole. The
instrumental applicative construction serves the first function, while the
locative applicative construction can perform both functions.20
Based on the discussion so far, the UV pattern should be deemed the
default pattern in Amis, which is proven by the fact that it is the unmarked
voice of the applicative constructions even when the UV markers do not
show up. Although the UV pattern enjoys unmarked status in Amis, there
are some predicates that seem to take the AV pattern by default, and for
such predicates, UV appears to be the marked pattern. Such predicates can
be illustrated by the pa- verbs. Consider the following examples of a pa- verb
plus the volitative mood suffix -aw:
Recall that in the previous discussion, I showed that when a root form is
suffixed with -aw, it takes the UV pattern. As we can see in (26a), the pa-
predicates appear with the AV case-marking pattern (i.e. nominative-dative).
However, when they are suffixed with -aw, the case-marking pattern
becomes the UV pattern. In other words, the pa- predicates behave like a
bare root form in the volitative mood construction. When the volitative
form pa-nanom-aw is prefixed with mi-, the case pattern becomes the AV
pattern again. These examples show that, unlike mi-, pa- does not have a
voice-marking function. However, pa- verbs follow the AV pattern by default.
To make pa- verbs appear in the UV pattern, the plain UV markers or the
applicative forms have to be used. Morphologically, the AV pattern appears
to be the default pattern of pa- verbs, while the UV pattern is a marked one.
However, syntactically, the UV forms actually turn a marked pattern (i.e. AV)
into an unmarked one. This may explain why the UV form pa- . . . -en and the
applicative form pa- . . . -an are found much more frequently than the plain
pa- forms in Amis.21
As mentioned, AV constructions perform both PSA-modulation and
argument-modulation functions. How about UV constructions? Though
appearing to be the basic pattern of Amis based on the case marking and
the default voice choice of the applicative constructions, they turn out to
be the marked voice choice for some predicates that usually appear with
the AV pattern by default. For such predicates, their plain UV constructions
perform a PSA-modulation function but no argument-modulation function
as the macrorole arguments and the NMR core arguments remain
unaffected. See Wu (2006a) for more discussion. However, the applicative
UV constructions display both functions, as now an adjunct becomes both
a macrorole and the PSA. The above discussion shows that both the actor
voice and the undergoer voice are deemed basic voice forms.22 Therefore,
Amis exhibits a split system in verbal morphology in spite of displaying
ergative features in the case-marking system and in some grammatical
constructions.
‘diligent’). In the RRG analysis proposed here, the voice marker that
appears on one-place predicates is treated as neutral voice as the only
macrorole of such verbs is not specified. This analysis is rather different
from the previous studies, which seem to assume a unified semantic role
for the single argument of intransitive predicates. For example, they either
label intransitive predicates as AV verbs (e.g. Liu 1999) or claim that the
single argument of intransitive verbs is patient (e.g. Chen 1987). In fact,
treating one-place predicates as AV or AF (i.e. agent-focus) verbs has been a
fairly common practice in the studies of other Formosan languages as well
(e.g. Atayal, as seen in Huang (2000) and Yeh (2015), and Tsou in Zeitoun
(1993), to name just a few). However, this unified-semantic-role approach is
implausible if we consider the rather different morphosyntactic behav-
iours of the single argument of the one-place predicates in Amis. In the
following, I am going to show how the RRG non-unified macrorole
approach for one-place predicates (Van Valin and LaPolla 1997; Van Valin
2005) can account for their peculiar morphosyntactic behaviours in two
constructions: the UV -en verbs and the evaluative construction.
a. T<om>ireng cingra.
<neut>stand 3s.nom
‘He is standing.’
a′. Tireng-en¼ako pa-kimad, ta paka-nengneng
stand-uv¼1s.gen caus-speech so.that ablt-see
kamo.
2p.nom
‘I will stand up when making a speech so that you can see (me) clearly.’
b. Ma-toni’ k-o-ni a titi.
neut-soft nom-cn-this lnk meat
‘This meat is soft.’
b′. Toni’-en¼ako k-o-ni a titi.
soft-uv¼1s.gen nom-cn-this lnk meat
‘I will soften this meat.’
For two-place or three-place -en verbs, all three of the rules in (29) are
applicable. But for the one-place -en verbs, only (29a) and (29c) will apply,
as there is only one macrorole in such verbs. The examples in (30) illustrate
how the rules in (28) and (29) work in Amis.
The Amis sentences in (31) and (32) all begin with a state predicate (i.e. the
evaluative predicate), which is then followed by the linker a and a UV verb. The
difference between the two types of evaluative construction is that the nom-
inative argument is a core argument shared by both predicates in (31) while the
one in (32) is an argument of the second verb but not the first one. Here we will
limit our discussion to the first type. A similar construction has been found in
other Formosan languages such as Seediq (Tsukida 2005) and Atayal (Yeh 2015).
Atayal examples are given in (33) (Yeh 2015: 132, original gloss).
Table 26.2 The relationship between the juxtaposed verbs in the BLAQ evaluative
construction, SVCs and commentative CCCs in Squlip Atayal: a comparison (adapted
from Yeh 2015: 148)
(35) Sharing of TAM information in SVCs and the evaluative constructions in Amis
Second, only the first verb in the named constructions can be conjugated
with the non-finite markers when following the negator caay ‘not’.25
(36) The negative sentences of SVCs and the evaluative constructions in Amis
However, unlike the SVCs in (34), it is the second verb in the evaluative
construction that controls the case-marking pattern of the sentence, but
not the first verb. As shown in (34c), even though the second verb appears
in the AV form, the arguments in the linked core are case-marked with
the UV pattern, following the first verb. In other words, the second verb
in the SVCs has no voice function. Nevertheless, the arguments of the
linked core in the Amis evaluative constructions always follow the case-
marking pattern of the UV verb, as seen in the comparison between (37a)
and (37b).
(37) The case-marking pattern of the Amis evaluative construction
Yeh (2015) in particular explores the answer to the question of why the
second verb in the evaluative construction has to be marked by the UV form,
but not the AV. She proposes that it is the undergoer that is evaluated in the
construction, and this pragmatic function accounts for the UV marking of
the second verb. However, the RRG approach might offer an even more
straightforward answer.
Following the RRG analysis, the first verb in (31) will have an undergoer
instead of an actor, and the voice sequence in the Amis evaluative con-
struction should be a neutral voice followed by a UV verb. The shared
arguments in such sentences play the role of undergoer (undergoer of
the intransitive verb and undergoer of a UV verb). Hence, it is quite
reasonable to maintain the syntactic status of the undergoer in both
cores. If the second verb takes the actor voice, the macrorole status of
the undergoer will be removed, as we have seen in the function of the
voice discussed earlier. The seeming paradox in Yeh’s study lies in her
treatment of the only argument of the first verb as an actor or a unified
thematic role, but the RRG perspective can provide an account that
resolves the paradox.
As for the SVC examples, the shared argument might play different
semantic roles in the two linked verbs, but this role is always an actor in
the second verb. This may explain why the second verb bears AV marking
though the AV does not have any voice function.
References
Wu, Joy Jing-lan. 2003. Clausal modifiers in Amis. Concentric: Studies in English
Literature and Linguistics 29(2): 59–81.
Wu, Joy Jing-lan. 2006a. Verb Classification, Case Marking, and Grammatical
Relations in Amis. PhD dissertation, University at Buffalo (SUNY).
Wu, Joy Jing-lan. 2006b. The analysis of pa- verbs in Amis. In Henry Y.
Chang, Lillian M. Huang and Dah-an Ho (eds.), Streams Converging into an
Ocean: Festschrift in Honor of Professor Paul Jen-kuei Li on his 70th Birthday
(Language and Linguistics Monograph Series Number W-5). Taipei: Aca-
demia Sinica.
Yeh, Maya Yu-ting. 2015. An atypical SVC? A study of BLAQ UV qu’ NP
construction in Atayal. In Elizabeth Zeitoun, Stacy F. Teng and Joy J. Wu
(eds.), New Advances in Formosan Linguistics, 131–155. Canberra: Asia-Pacific
Linguistics.
Zeitoun, Elizabeth. 1993. A semantic study of Tsou case markers. Bulletin of
the Institute of History and Philology 64(4): 969–989.
Notes
1 www.apc.gov.tw/portal/docList.html?CID¼7AD36169AB07E1D0.
2 The phonetic symbols used in the transcription generally follow the IPA
system, with the following exceptions: /e/ stands for schwa [ ], /d/ for e
voiceless lateral [ɬ], /’/ for epiglottis stop /ʡ/, and /ng/ for /ŋ/. The glosses in
general follow Wu (2006a), which adopts some of Liu’s (1999) morphemic
analyses (e.g. the separation of noun classifiers and case markers) with
some slight modification.
3 There are two classes of nouns identified in Amis: personal proper nouns,
including people’s names and kinship terms, and common nouns. The
noun class marker usually appears with the case marker and forms a
case-marking composite, but the case marker does not show up in
predicate position.
4 The analysis of the prefixes ka- and pi- as markers of non-finite forms
follows Bril (2016), but Bril specifies pi- as a non-finite marker for actor
voice verbs (i.e. NFIN.AV). Here, pi- is simply glossed as NFIN.
5 However, Wu employs an ideophone-forming construction and classifies
the root forms into five classes: object, state (attribute), state (transient/
result), activity, achievement/semelfactive (Wu 2006a: 156).
6 As argued in Wu (2006a), the voice markers carry both inflectional and
derivational function, and the latter function changes not only the
lexical categories of the attached roots or stems but also their semantics.
These voice markers also have their own decomposed structures. For
example, a UV verb does not merely enhance the status of a patient role
as it would in an analogous English passive sentence. It also has a
different meaning due to the UV marker ma- and the derived verb might
become an (active) accomplishment verb.
7 The factual marker -ay also carries a nominalizing function, which
roughly means ‘someone who . . .’ or ‘something which . . .’.
8 Exceptions are found in state predicates with a gradable feature (e.g. ma-
lalok ‘diligent’); they can readily appear after a case marker in a compara-
tive construction. See Wu (2003, 2006a) for details.
9 Examples (8b) and (8c) exemplify a nominal type of wh-question . Only
the PSA can be the missing argument in the nominal clause. This will be
further discussed in Section 26.5 on grammatical relations.
10 For a special type of three-place predicate that is prefixed by pa-pi-,
Principle B is the only one that is applied.
11 To simplify the discussion, the LS of ma- (active accomplishment, UV) is
not represented in the LS of the ma- UV construction of the three-
place predicates.
12 The semantic role of the noun marked by the dative case is indicated in
parentheses after the translation.
13 The differences between the plain UV and the applicative UV construc-
tions are subtle. In general, the -en plain UV construction emphasizes the
volition of the actor, and it is more likely to refer to a future event; it is
also the form that is used in imperative sentences. The ma- plain UV
construction highlights the affectedness of the undergoer. The mi- . . . -an
applicative UV stresses the identity of the actor.
14 The to RP in (12a) in fact manifests an external causer for the event
described by the predicate, though the predicate is non-causative. The
same phenomenon is also found in the to RP in (12c). This explains why
they can serve as the argument for the mi- counterparts in (13), which
carry a causative reading after derivation. Functionally speaking, the to
marker here is similar to the English preposition from, which appears to
be causative in its predicative roles (Jolly 1993: 293), as in the sentence
John died from Malaria. The causative version of this English sentence
would be Malaria killed John.
15 This set of rules cannot work for the verbs suffixed by the UV marker -en.
We will therefore propose another set of rules for such verbs in the
special discussion of Amis one-place predicates in a later section of
the chapter.
16 The RC is boldfaced in the examples. The ‘__’ and the coreferencing
subscript in the examples is for expository purposes only. It does not
refer to a trace or a null copy in the site of the missing RP. RRG does not
posit such notions.
17 Choosing one type over the other crucially depends on the status of
the displaced RP. If it is a core argument, then the nominal type is
preferred or even required for some speakers; if it is an oblique
argument or adjunct, then the verbal type is allowed. As only the